x-xEPA
            United States
            Environmental Pro tec lion
            Agency
            Office of
            Noise Abatement and Control
            Washington, D.C. 20460
EPA 550/9-79-102
November 1979
            Noise
Comparison of Various
Methods for Predicting
the Loudness
and Acceptability of Noise
            Part II:
            Effects of Spectral
            Pattern and Tonal
            Components
                                  LOUDNESS
                                   LEVEL (PHON)

-------
COMPARISON OF VARIOUS METHODS
FOR PREDICTING THE LOUDNESS AND
      ACCEPTABILITY OF NOISE

                 Part II

   EFFECTS OF SPECTRAL PATTERN
      AND TONAL COMPONENTS
               November 1979


                Prepared by

            B. Scharf and R. Hellman
          Auditory Perception Laboratory
            Northeastern University
          Boston, Massachusetts 02115
                Prepared for

       U.S. Environmental Protection Agency
       Office of Noise Abatement and Control
            Washington, D.C. 20460

-------
                                  PREFACE
      The United States Environmental Protection Agency (EPA) was charged
by Congress in the Noise Control Act of 1972, as amended by the Quiet
Communities Act of 1978, to conduct or finance research to investigate
"...the psychological and physiological effects of noise on humans and the
effects of noise on domestic animals, wildlife, and property, and the deter-
mination of dose/response relationships suitable for use in decision making..."
(Section 14(b)(l)).
      Pursuant to and as part of this mandate, EPA has undertaken investi-
gations to determine and quantify subjective reactions of individuals and
communities to different noise environments and sources of noise.  A specific
series of studies has been initiated to determine the best methods for eval-
uating subjective magnitude and aversiveness to noise on the basis of spectral
and temporal properties, and to ascertain the importance of and means for
including nonacoustical factors in the evaluation of general aversion to noise.
The overall purpose of this line of research.is to derive a more solid basis
for assessing the aversiveness of noise and the benefits of noise control.
The program calls for detailed analysis and evaluation of available data from
both the laboratory and the field to assess the relative validity and pre-
dictiveness of various subjective acoustic ratings (spectral weightings and
calculation schemes), as well as to acquire new data where appropriate.
      Findings have been published previously in EPA Report No. 550/9-77-101
entitled "Comparison of Various Methods for Predicting the Loudness and
Acceptability of Noise."  That report dealt with the ability of commonly
employed frequency weightings and calculation schemes to predict and quantify
subjective aspects of sound.   The results of the study showed the calculation
schemes to be superior in predictive capability to the frequency weightings.

                                    111

-------
The D- and E-frequency weightings were significantly better than the B- and
C-weightings.  The A-weighting was only slightly more variable than the D-
and E-weightings.  All frequency weightings were level dependent with the
predictive capability worse at higher levels.  Analysis of the results with
regard to the type of noise and the presence of tonal components was not
conclusive due to a limited amount of available data.
      The purpose of the investigation described in this report was to under-
take a more detailed, rigorous, and systematic analysis of the previously
compiled psychoacoustic data in order to (a) account for certain apparent
anomalies in the data analyzed earlier as part of this program, (b) examine
the sensitivity of various frequency weightings and rating schemes to spectral
differences of the sound stimuli used in the investigations, and (c) evaluate
subjective response to discrete frequency components superimposed over a back-
ground.  The results  provide partial but needed information on the relative
ability of computational procedures and frequency weightings to assess sub-
jective loudness and  acceptability of sounds with different spectral shapes,
the necessity of tonal corrections at low and high levels of noise, an indi-
cation as to the magnitude of a correction, and the overall effectiveness of
commonly used tonal correction procedures.
      EPA believes that further evaluation of data on the subjective effects
of noise will foster  the development of techniques to demonstrate additional
benefits of noise control beyond that exhibited by currently used procedures.
Fulfillment of this objective awaits further study within this series.  The
results published in  this report, however, do provide an important step to-
ward a more complete understanding of the phenomena of human subjective
response to noise.

OFFICE OF THE SCIENTIFIC ASSISTANT
  TO THE DEPUTY ASSISTANT ADMINISTRATOR
OFFICE OF NOISE ABATEMENT AND CONTROL
U.S. ENVIRONMENTAL PROTECTION AGENCY
                                         iv

-------
                                   Abstract




     The present report is a continuation of an earlier report by Scharf,




Hellman, and Bauer (1977).  The objectives are (1) to determine whether sub-




jective judgments of particular types of noise, categorized by spectral




shape, are better approximated by  some descriptors (frequency weightings and




calculation procedures) than by others, and (2) to investigate the role of




tonal components in these studies  and to assess the adequacy of several




tone-correction procedures.  The analysis of data by spectral shape produced




a mixed outcome.  Results showed that no overall advantage would accrue




from regrouping sets of data across studies on the basis of similar spectral




shapes.  However, although variability was not reduced when considered




across nine spectral categories, the interaction between spectral shape




and descriptor was highly significant (p < .001).  The examination of over




500 spectra with and without tonal components provided only tentative




support for the trends noted in the literature.  When the judged attribute is




either loudness or noisiness, tonal components do not seem to add to the




subjective magnitude of broad-band noise below 80 dB sound pressure level.




At higher levels, according to one large-scale study, tonal components




seemed to add the equivalent of 2  dB to the judged noisiness.  No data could




be located that would permit adequate assessment of the contribution of tonal




components to the "absolute" magnitude of judged annoyance or unacceptability




(as distinct from noisiness or loudness).  Given the small effect of tonal




components in the present group of studies, the evaluation of three different




tone-correction procedures (FAR 36, 1969; Kryter and Pearson's, 1965;

-------
and Stevens's, 1970) could not lead to definitive conclusions about their




relative merits.   Although a small correction may be necessary for the




presence of tonal components at high levels, the tone-correction procedures




now available cannot be properly evaluated until more appropriate data that




demonstrate the need for a tone correction are obtained.
                                     VI

-------
                         ACKNOWLEDGEMENTS









     We thank Professor Merv Lynch for his  aid  in statistical  analyses,




Barbara Kane for implementing the ANOVA programs  on  the  computer,  and




Harvey Branscomb for writing programs  to handle the  various  tone-correc-




tion procedures.  We also wish to express our thanks to  Jeffrey Goldstein,




project director at EPA, for his  constructive reviews  of initial drafts of




this report.  A number of undergraduate and graduate students  at North-




eastern University also helped us with the  many details  of this report;




they include Eleanor Arpino,  Angela Ashton,  Maureen  Hogan, Tom Horton,




and Patricia Moran.  Correspondence and most  of the  typing were beauti-




fully handled by Ana Silfer.
                               VII

-------
vm

-------
                             TABLE OF CONTENTS



                                                                       Page

I.     INTRODUCTION                                                       1

II.   SPECTRAL SHAPE                                                     3

III.  TONAL COMPONENTS                                                  21

      1.  Composition of Studies with Respect to Tonal Components       24

      2.  Evidence Demonstrating a Need for a Tone Correction           24

      3.  Descriptions of Tone-Correction Procedures                    37

          a)  PNLC or FAR 36 Tone Corrections                           37
          b)  Kryter and Pearsons°s (1965) Tone-Correction Procedure    38
          c)  Stevens's (1970) Preliminary Tone-Correction Procedure    40

      4.  Other Tone-Correction Procedures                              44

      5.  Evaluation of Tone-Correction Procedures                      45

          a)  Variability                                               45
          b)  Mean Differences Between Calculated and Observed Levels   48

      6.  Summary of Findings Relative to Tonal Components              52

IV.   CONCLUSIONS AND RECOMMENDATIONS                                   53



APPENDICES

A.     CATEGORICAL ANALYSIS ACCORDING TO SPECTRAL TYPE                  A-l

B.     "ANOMALOUS" DATA                                                 B-l

C.     STEVENS'S TONE-CORRECTION - 1970 PRELIMINARY PROPOSAL            C-l

D.     ERRATA AND ADDENDA TO SCHARF, ET. AL. (1977)                     D-l

R.     REFERENCES                                                       R-l

-------
                                I.  INTRODUCTION

     A recent report by Scharf, Hellman, and Bauer (1977) examined data from

23 studies  in which subjects had judged the subjective magnitude of a large

variety of  noises.  The aim of  the Scharf, et.  al. (1977) investigation was

to determine how well various frequency weightings (presently incorporated

or proposed for use on sound level meters) and calculation procedures assess

the  subjective magnitude of noise.  One important conclusion, based on a total

of over 600 spectra, was that the calculation procedures predicted subjective

magnitude with less variability* and with greater validity** than did the

frequency weightings.

     Among  the six frequency weightings studied, the B- and C-weightings

were the poorest predictors of  subjective magnitude while the D1-, D2-,

and E-weightings were the best  predictive weighting functions.  It was also

noted that  the A-weighting was  less than 0.5 dB more variable than the D1-,

D2-, and E-weightings.  Among the five calculation procedures studied,

Stevens's Mark VI (1961), Mark  VII (1972), and Zwicker's (1958)  loudness

calculation procedures were the least variable, but Perceived Noise Level

(Kryter 1959) was almost as reliable.  Tone-corrected Perceived  Noise Level

(following  the FAR 36 procedure, 1969) was a somewhat poorer predictor.   Mark VI

and Perceived Noise Level yielded the calculated values that were closest,  on

the average, to the observed or judged values,  although all of the frequency

weightings  and computational procedures examined were about equally variable

in this respect.
 *The index of variability was the standard deviation of the calculated levels
  of a group of sounds judged subjectively equal or the standard deviation of
  differences between calculated and judged levels.  These typically ranged
  from 2 to 4 dB.

**The calculation procedures yielded an absolute calculated level closer to
  the observed level.

                                      1

-------
     The objectives of the present investigation are (1) to determine




whether subjective judgments of particular types of noise, categorized




by spectral shape, are better approximated by some descriptors (frequency




weightings and calculation procedures)  than by others,  and (2) to investigate




the role of tonal components in these studies and to examine the relevancy




of existing tone-correction procedures.




     Each of these aims is addressed  separately with overall results and




conclusions provided in Section IV.   Appendices A,  B, and  C include more




detailed analyses.

-------
                               II. SPECTRAL SHAPE




     Particular types of noise often have distinctive spectral characteristics




such as the low-frequency spikes of transformer noise,  the low-frequency




emphasis of vehicular noise, the mid-frequency bulge of many machine noises,




the high frequencies of an electric bell, and so forth.  In the earlier report




(Scharf, et. al. , 1977, Table V), twenty of the studies examined were classi-




fied according to the specific source or type of noise.  The six sources




considered were aircraft, industrial, vehicular, and household, as well as




artificial and miscellaneous noises.  A statistical analysis of the differ-




ences among the data for these six noise sources was performed in the present




study.  For purposes of this analysis, the vehicular noise category, for




which there was only one set of data, was combined with the aircraft noise




category to form a general transportation noise group,  thus yielding a total of




five source types.  As shown in Table I, a partially hierarchical analysis of




variance (ANOVA, Winer, 1962) revealed no significant differences in the




predictive ability of the ten descriptors among the five source types.




However, the interaction between source type and descriptor was significant




(p < .01).  Despite the statistical significance of this interaction, the




differences among the descriptors are too small to provide a basis for




concluding that certain types of noises are better assessed in any meaningful,




practical sense by one particular descriptor than by another.  Moreover, the




small number of studies contained within each source type category indicates




that noise source type and study are confounded.

-------
                                   Table I
               Summary Table for Partially Hierarchical ANOVA;
         Five Source Types by Ten Descriptors (PNLC Has Been Omitted)
Source of Variance Sum of squares
Source type
Between groups
(error term)
Descriptor
Descriptor by source
Within groups
50
175

63
36
97
.92
.20

.05
.51
.60
Degrees of Mean
Freedom Square
4
22

9
36
198
12.73
7.96

7.00
1.01
.49
F P
1.60 NS


14.21 <.001
2.06 <.01

(error term)

-------
     The  lack of  significant  differences  among source types may, in fact,

be attributable to  the  rather gross  classification scheme whereby a

wide variety of spectral  shapes were included within each source type.

Thus, this  analysis may have  obscured real differences among spectral

shapes.   A  more homogeneous classification can be achieved by regrouping

spectra from different  studies according  to spectral type or shape.  It is

possible  that for certain spectral shapes, particular descriptors (frequency

weightings  or calculation procedures) predict subjective judgments better

than other  descriptors. If so, descriptors could then each be applied, in prac-

tice, to  those spectral shapes to which they are best suited.  Accordingly, each

noise spectrum within the 19  studies  listed in Table II of Scharf,  et. al.

(1977)* was placed  into one of nine  spectral categories:  (1) negative slope,

(2) positive slope, (3) broadband and flat, (4) narrow band, (5) U-shaped,

(6) inverted U-shaped,  (7) low-frequency  peaks or valleys, (8) mid-to-high

frequency peaks or valleys, and (9) mixed peaks or valleys.   Figures 1 to 9

provide examples  of sound spectra from each of the nine main categories.   The

spectra represent noises  from both artificial and natural noise sources.

(Appendix A gives more  detailed definitions of the spectral shapes and a more

detailed  breakdown within the main spectral categories.)

     Table  II presents  the standard deviations (SDs) averaged across the nine

spectral  categories for (1) those sets of subjective data that did not provide

judged loudness levels,  (2) those that did, and (3) all spectra combined.**
 *The data by Pearsons, et.  al.  (1968) were not included in this analysis.  Wells
  300 and Wells 400 are counted as one study.

**The SDs for each spectral category are provided in Tables A-2 and A-7 of
  Appendix A.  Note that the means of the SDs were computed without regard
  to the number of SDs contributed by each category.

-------
   o
   o
00

T3
O
CD
    o _;
    o
    •a-
    o _;
    o
    CM
    o _;
                                                 BERGLUND,ET AL 1976 PILEDRIVER
      10
                          n       i     i   i  r  i MM


                          100                      1,000


                                   Frequency (Hz)
ir  i  riii


          10,000
           Figure 1.  Spectral Category  1A (Strong Negative Slope)

-------
   S-,
   O _I
   o
   co
   o _;
s   s
   s-=
o  o _
    O .
    ro
   o
   CN
I  I  I  I I  11	


        1,000


     Frequency (Hz)
                                                                        PEARSON - BENNETT TEST 1 STI M 2 AW


                                                                      i   i  i ii 11	r—r
      10
i—r
                              100
                                                            10,000
                              Figure 2.   Spectral  Category 2  (Positive Slope)

-------
00
                o
                o
               O
               CO
            —   o
            m   to
                8 -E

            C/3
               O
               n
               o
               CM
                  10
                                                                                   LUBCKE, MITTAG, PORT - THIRD OCTAVE
                              I    I  I  I  I I I
100
!   I  I  I  I  I I        I



           1,000



       Frequency (Hz)
I   I  I  I  I III



          10,000
                                          Figure 3.   Spectral  Category  3  (Broadband  Flat)

-------
     o
     o
• o
 CO
     o
     r-.
     o
     (O
     o
     in
     o _
     o
     o
     o
     CM
     O 	
                                                         KRYTER- PEARSON 1963 N - 3
       10
                I    I   I  I  I  I I  I I



                                100
I     I   I  I  I  I I I



                1,000



Frequency (Hz)
I    I   I   I  I I  I I



               10,000
                 Figure 4.  Spectral Category 4 (Narrow-band  Noise)

-------
                                                                                  SPIEGEL N20
                                                                                THIRD OCTAVE
                                                                                    (64 PHONS)
10
                         Figure 5.  Spectral Category 5 (U-Shaped)

-------
o
00
o _^
o _i
o _;
CM ~
                                           PEARSON-BENNETT TEST 1 STIM 18 DX
               I   r  i i  i  i i
   10
100
i  i  i i i


      1,000
i   i  i  i  i i  11


          10,000
                                 Frequency (Hz)
       Figure 6.   Spectral  Category 6  (inverted U-shaped)
                                 11

-------
    8  -,
    o
    en
    o
    oo
    o
    CD
2

CD

3
T3

i
V)
    O
    CO
    o
    CN
                                                                                        WELLS UHV NOISE 8
       10
                   I   I  I  I  I I  I I


                               100
1    I   I  I  I  I I  I        I     I   I  I  I  I I I I


                1,000                    10,000


             Frequency (Hz)
1    I   1  T I  I I  I
                       Figure 7.   Spectral  Category  7  (Low-Frequency Peaks  and Valleys)

-------
u>
           8 _
           o
           CO
       o
       in
           o
           (O
           o
           CN
              10
                                                                                FISHKEN N17 500 + 2000 HZ AT +15 DB


                                                                              1  I  I  I I I |       I    I   I   I I  I  I I |
                	1	1	1  I  I  I M|	1                 (


              100                      1,000                    10,000



                                  Frequency (Hz)



Figure  8.   Spectral  Category 8  (Mid-to-High  Frequency Peaks or Valleys)

-------
    o
    o .
    i-
    s-
CO
-a
£   S'
Q.
o
CSS
                                                                     WELLS 1970 AIRCRAFT NOISE 12  3-3-77
     10
                  I   I   I  I I  I I
                              100
I   I   I  I I  I I

            1,000

         Frequency (Hz)
I    I   I   I  I  I I I I


               10,000
I    I    I  I  I I  II
                         Figure 9.   Spectral  Category  9  (Mixed Peaks or Valleys)

-------
                                  Table II
                 Variability of Calculated Levels of Noises
                  Grouped by Spectral Category or by Study

(Standard deviations in decibels computed either from the calculated levels of
a group of sounds judged subjectively equal or from the differences between
calculated and judged levels (loudness levels).  The smaller the standard
deviation, the closer the scheme comes to predicting the measured subjective
equality of a set of sounds.)
SOURCE                     N/n
Spectral categories
(Based on calculated
levels)
Spectral categories       335/56
(loudness levels only)

Spectral categories       633/90
(total)
                   Descriptors

                Dl    D2    E
            VI
VII   PNL   ZWI
298/34   2.7   2.2   2.3   2.1
         2.8   2.6
Grouped by study (total)  763/28   3.1
from Scharf, et. al.
(1977) Table II
corrected
            1.9
2.1   2.2
2.7   2.6   2.2   2.3
      2.7
               2.7   2.7    2.6    2.3    2.2    2.6
2.7
         2.9   3.0   3.0   3.0    2.4    2.4    3.1    2.3
2.5
                              2.4
LEGEND:

N  = number of spectra

n  = number of standard deviations

A  = standard sound-level meter
     weighting

Dl = sound-level meter weighting,
     better known as D, adopted
     by International Electro-
     Technical Commission (1975).

D2 = sound-level meter weighting
     proposed by Kryter, K.D.
     (1970), Table 2.
            E  = sound-level  meter  weighting  proposed
                 by Stevens  (1972)  and  circulated  as
                 ANSI Draft  document  Sl.XX/104

            Mark VI = ANSI  S  3.4 (R1972)  procedure
                      for the computation of  loudness
                      of noise.

            Mark VII = proposed  by  Stevens  (1972)

            PNL = perceived  noise level

            ZWI = based on  Zwicker  (1958).  Computer
                  program from Paulus and Zwicker
                  (1972)
                                      15

-------
Table II also presents the SDs previously calculated across studies (Scharf,




et_._ al^_, 1977, Table II, with the minor corrections given in Appendix D of this




report).




     As shown in Table II, most of the sounds represented in row 1 of Table




II were judged with respect to some evaluative attribute such as noisiness,




unacceptability, etc., whereas the sounds represented in row 2 were judged




only with respect to loudness.  Thus, the data contained in rows 1 and 2




demonstrate the same tendency noted in Scharf, et.  al.  (1977, Table V).  Those




data showed that the studies in which loudness was  judged yielded larger SDs




than those studies in which an evaluative attribute other than loudness was




judged.  The difference, however, was not statistically significant.  The most




probable basis for the difference, described in detail  in Appendix B, is the




wider range of levels covered by the loudness studies than by those studies in




which an evaluative, attribute was judged.




     The most revealing comparison in Table II is between overall SDs calculated




across spectral categories (row 3) and those calculated across studies (row 4).




Except for the A-weighting, paired SDs do not differ between rows 3 and 4 by




more than 0.1 dB.  Thus, classifying the spectra according to shape does not




reduce overall variability.  Underlying this analysis was the assumption that




a descriptor would be less variable if applied to groups of spectra of the




same shape than to groups of spectra of different shapes.  Although variability




is not reduced when calculated across all the spectral categories, it may be




smaller for particular descriptors applied to particular spectral categories.




The interaction between category and procedure is considered in Table III.
                                      16

-------
                                 Table III

             Summary Table for Partially Hierarchical ANOVA:
               Nine Spectral Categories by Ten Descriptors
Source of Variance
Sum of squares
Degrees of    Mean
Freedom      Square
Source type
Between groups
Descriptor
Descriptor by source
Within groups
143.18
563.91
28.64
38.70
161.04
8
71
7
56
497
17.90
7.94
4.09
.69
.22
2.25

12.63
2.13

<.05

<.001
<.001

                                      17

-------
     Table III presents ANOVA for spectral type or shape by descriptor.




This analysis shows that (1) the differences among the nine spectral cate-




gories were significant at the .05 level, and (2) the differences among




the ten descriptors were statistically significant at the .001 level.




Further, the interaction between spectral shape and descriptor was signifi-




cant at the .001 level.  However, despite this significant interaction, a




meaningful multiple contrasts test could not be performed due to large




variations in numbers of spectra and in numbers of SDs among the nine cate-




gories .




     Also relevant to the analysis by spectral categories is the question




of differences between calculated and observed loudness levels.   Table




IV, based partly on Table A-10 in Appendix A and partly on Table IV in




Scharf, et. al.  (1977), gives the overall means of the mean differences for




over 300 noises grouped either by spectral category or by study.  The corre-




sponding SDs of the mean and total ranges are also shown.  Except for Zwicker's




loudness calculation procedure and Mark VII after the required addition




of an 8-dB constant, all the descriptors are more discrepant for the sounds




grouped by spectral category than for the sounds grouped by study.   Of




more importance, however, is the variability of the mean difference.  Both




the range and SDs are significantly smaller (p < .01 by t-test)  for the sounds




grouped by spectral category than for those grouped by study, with  the sole




exception of the SD for the A-weighting.  This decreased variability is




especially noteworthy since studies that differed with respect to procedures,




standards, and instructions were broken up and individual spectra assigned to




various spectral categories.  These methodological differences would be




expected to increase variability.  Since the opposite occurred,  it  is likely
                                     18

-------
                                  Table IV

                  Calculated Minus Observed Loudness Levels
                       (Mean Differences in Decibels)

(Overall means based upon differences for 335 spectra grouped according to
spectral type as per Table A-10 in Appendix A.  Overall means are also shown
for the same spectra when grouped by study as per Table IV of Scharf,  et
1977.)
By Spectral Category

Mean of Mean Differences   -12.1

S.D. of Means

Range


By Study

Mean of Mean Differences

S.D. of Means

Range
A
-12.1
4.8
16.0
-10.8
4.5
17.8
Dl
-5.3
4.0
12.6
-4.5
4.7
18.8
D2
-5.8
4.3
13.7
-5.0
4.8
19.3
E
-6.8
4.1
12.6
-6.2
4.6
17.7
VI
-1.2
3.2
11.6
-0.1
4.5
17.4
VII
-8.6
3.2
11.1
-6.9
4.3
15.3
PNL
-1.4
3.1
10.8
-0.0
4.7
19.3
ZWI
3.1
3.0
8.8
5.1
4.2
14.7
                                      19

-------
that grouping by spectral shape meaningfully enhances the validity of




the descriptors.  Moreover,  the four calculation procedures with their much




greater flexibility showed a larger drop in variability than did the weight-




ing functions.  In contrast, the A-weighting with its strong deemphasis of low




frequencies revealed an increase in the standard deviation. As can be seen in




Table A-10, the A-weighting grossly underestimated the level of sounds with




much energy in the low frequencies and less grossly underestimated spectra




with little energy in the low frequencies.   To a lesser extent, the other




frequency weightings also deemphasize low frequencies, and this deemphasis




becomes detrimental at high levels (Scharf, et.  al.,  1977, Figures 6 to




8).




     Furthermore,  it should be pointed out  that  categories 7, 8 and 9 which are




distinguished by the presence of low-frequency spectral peaks or valleys,




mid-to-high-frequency peaks or valleys, and mixed peaks or valleys, respec-




tively, include many sounds wJ^h tonal components.  Defined as projecting at




least 3 dB above their neighboring third-octave  bands (see Section III), tonal




components were identified in over 80 percent of the  sounds in categories




7 and 8, and in 30 percent of those in category  9.  The SDs are presented in




Tables A-12 and A-13 of Appendix A.  No clearcut differences were found between




those spectra with tones and those without  (except for part of category 9, as




discussed in Appendix A).  The general problem of tonal components is treated




next, in Section III.
                                      20

-------
                           III. TONAL COMPONENTS




     A number of  studies have  reported that people react more negatively




to noises containing tonal components than to the same or similar noises




without tonal components.  Tonal components appear to add more to the unpleasant-




ness of a noise than the same  amount  of acoustical energy would add if




spread over a wide band of frequencies.  Reports in the literature (Copeland,




1960; Hargest and Pinker, 1967; Kryter and Pearsons, 1965; Little, 1961;




Little and Mabry, 1969; Pearsons, 1968; Pearsons and Bennett, 1969, 1971;




Pearsons, Bishop  and Horonjeff, 1969; Pearsons and Wells, 1968, 1969;




Wells, 1967, 1969b) show that  tonal components add the equivalent of from 2




to 15 dB or more  to the annoyance of  a sound than would be expected from the




increase in overall energy.   Several  reports show that loudness or noisiness,




as distinct from  annoyance or  objectionability, is not affected by the




presence of tonal components  (Fishken, 1971; Kryter and Pearsons, 1963;




Rule, 1964; Rule  and Little,  1963).   One report (Niese, 1965) showed that




tonal components  affected both loudness and annoyance to the same degree.




Another report (Goulet and Northwood, 1972) found no effect of tonal compon-




ents on either loudness or annoyance.  In both these studies stimuli were




presented at levels between 45 and 75 dB sound pressure level.  On the other




hand, the investigations showing that tonal components do contribute unduly




to annoyance were conducted mostly at levels of 85 dB and higher.




     The present report evaluates a number of the studies cited above.   Some




studied sounds with tonal components  artificially added (Fishken, 1971;




Pearsons and Wells,  1969; Wells,  1969b),  and others studied natural sounds




that  contained tonal components (Pearsons and Bennett, 1969, 1971; Wells,
                                      21

-------
1970, 1972).   Although many studies not cited above,  but examined in Scharf,




et.  al.  (1977),  did include some noises with tonal components,  such noises




did not  usually constitute a large part of a given study.   Nevertheless, to




provide  a preliminary analysis of the effect of pure  tones on judgments of




loudness and annoyance, 27 of the 28 sets of SDs from Scharf, et. al.  (1977)




were divided into two groups.*  One group of 12 SDs was obtained from subjective




judgments of spectra without tonal components,  and a  second group of 15 SDs




was obtained from subjective judgments produced by spectra that contained




tonal components.  The presence of tonal components was based on the respective




authors' definitions.  The results are found in Scharf, et. al. (1977),




Table V.




     A partially hierarchical ANOVA (Lynch and Huntsberger, 1976) based on




those data revealed no significant difference between the SDs for 10 of the




11 descriptors (PNL tone corrected in accordance with FAR 36 was omitted).




The interaction between the presence or absence of tonal components and




descriptors was also not significant.  This negative  finding, however, may




not be meaningful.  First, many of the studies included within the group




without  tonal components had a few spectra with components.  Second, other




differences (such as attribute judged) among studies  could have obscured




any effects of tonal components on the variability of the descriptors.




Tiird, and most important, is that if the effect of tonal components is to




increase the unpleasantness of a sound, then sounds all or most of which




contained tonal components would all be more or less  equally affected.  Most




of the descriptors would then show no change in their variability unless




"absolute" levels were measured.  Such levels were not measured in most of
*The data by Robinson and Bowsher (1961) were not included in this analysis,
                                      22

-------
the studies involving sounds with tonal components; sounds were usually all judged

equal to a standard, and hence only a measure    variability was meaningful.

     For the present report, a detailed analysis of more than 600 spectra*

from Scharf, et. al., (1977) was undertaken to identify those spectra that

contained tonal components.  The criterion for identification of a tonal

component was that  a third-octave band must have a level at least 4.75

dB above that of either of the immediately adjacent third-octave bands.

This criterion was  adopted to assure that the tone is at least 3 dB above

the noise in the band of interest, and is similar to the FAR 36 procedure.**

If the 4.75 dB criterion is exceeded, then the tone in the given third-octave

band must be at least 3 dB above the level of the noise in the band that

contains it.  It was felt that, rather than rely on the authors' definition

of tones which may vary among authors, a precise identification of the

spectra containing  tonal components would permit a finer determination

of how well the different sound descriptors handle such stimuli.  (A partial

analysis of this type is presented in Appendix B for individual studies.)

     Several procedures specifically designed to "correct" for tonal compo-

nents will be evaluated in addition to the eight descriptors examined in

Section II.  These  include the FAR 36 (1969) procedure, which was identified

as PNLC in Scharf, et. al. (1977); a different correction to Perceived Noise

Level proposed by Kryter and Pearsons (1965); and a procedure tentatively

proposed by S. S. Stevens (1970) explicitly for use with Mark VII but appli-

cable to any of the other descriptors.  To augment the power of these analyses,

a large-scale study by Ollerhead (1971, 1973) has been added to the


     *The data by Pearsons, et. al. (1968) were not included in this analysis.

     **No distinction is made between a "true" tonal component and a sharp
       increase in  level over a restricted range of frequencies.

                                      23

-------
original group of studies examined in Scharf, et. al.  (1977).  Not  only  do




these additional 104 spectra include many stimuli with tonal components,  but  a




iudged  level for each of the stimuli is provided as well.







1.   Composition of Studies with Respect to Tonal Components




     More than 500 spectra with and without tonal components including 104




spectra from Ollerhead (1971, 1973) underlie the analysis described  in




this section.  Of approximately 300 spectra with tonal components,  over




one fourth contained more than one tone.  Most single components  fell between




500 and 2000 Hz; the remainder were nearly evenly divided between those  at




frequencies below 500 Hz and those above 2000 Hz.  With respect to  tone-to-




noise ratio, over half the components were less than 13 dB above  the surroun-




ding third-octave bands, one third were between 14 and 23 dB, and less than




one tenth were more than 23 dB above the noise.  Approximately half  the  tonal




components were at a sound pressure level between 60 and 80 dB, 30  percent




were above 80 dB, and 20 percent at 60 dB or lower.







2.   Evidence Demonstrating a Need for a Tone Correction




     As noted above, tonal components may contribute unduly to the  unpleasant-




ness of noise.   If so,  then those groups of noises that are a mixture of




sounds both with and without tonal components ought to show more variability




for a given descriptor  than either a group of noises all with tonal  compon-




ents or a group of noises all without.   Accordingly, the whole set  of noises




was first  examined for  this posited difference in variability without regard




to the attribute judged (whether loudness or some evaluative attribute),
                                      24

-------
 tone-to-noise ratio,  or overall  level,  parameters  which may  in  fact  be




 relevant  to the effect  of tonal  components  on human  response.




      Table  V presents the standard  deviations for  542  spectra from  13 studies




 and  subsets listed  in column 1,  Table  VI,  that had at  least  three spectra




 with tonal  components and at least  three without.  The mean  SDs  for  all  the




 sets of spectra,  both with and without  tonal  components,  are given  in the




 first row,  followed by the mean  SDs  for those spectra  with tonal components,




 and  then  by those without tonal  components.   The SD  of the SDs upon  which




 the  mean  values are based are also  shown.   For every descriptor  the  SD for




 the  overall group is  larger than the SD for either subgroup.  This result




 suggests  that sounds  with tonal  components  are judged  somewhat differently




 from sounds without;  that effect is  apparent  for this  analysis even  when




 studies that contained  relatively soft  sounds judged with respect to loudness




 are  included.




      However,  when  just those studies  are examined that involved evaluative




 judgments of annoyance, unacceptability, etc.  (and studies that  involved




 loudness  judgments  are  excluded), the  picture is altered.  Table VII shows




 that  in the annoyance studies, those spectra  with  tonal components produced




 the  largest SDs under all eight  descriptors,  while those  spectra without




 tonal  components  produced the smallest  SDs.   The presence of tonal components




made  the descriptors  more variable without  apparently  affecting the  SDs




obtained  for  the  mixture  of sounds both with  and without  tonal components.




Had  spectra with  tonal  components been  judged differently, on the average,




than  spectra without  tonal  components,  the  SDs  for the overall group would




have been increased,  not  decreased slightly as  they  are in Table VII.
                                   25

-------
                                   Table V

          Standard Deviations (In Decibels) for Spectra Both
          with and without Tonal Components, for Spectra with
          Tonal Components, and for Spectra without Tonal
          Components.  (Means were Unweighted.  Attribute Judged:
          Loudness, Annoyance, Noisiness, Etc.)
                   Number
                     of     Number*   Frequency Weighting   Calculation Procedure
	Spectra   of SDs    A    Dl    D2     E     VI    VII    PNL  ZWI

Mean SD (in decibels)
Spectra Both
with and without
components          542
Spectra with
tonal components    314

Spectra without
tonal components    205
29     3.1   3.0   3.1   2.9    2.6  2.7  2.8    2.7


29     2.6   2.4   2.4   2.3    2.1  2.1  2.4    2.3


20     2.7   2.4   2.6   2.3    2.1  2.1  2.2    2.4
SD of SDs (in decibels)
Spectra Both
with and without
tonal components

Spectra with
tonal components

Spectra without
tonal components
       1.2   1.2   1.2   1.2    1.1  1.4  1.0    1.2
       1.4   1.2   1.3   1.2 ,   1.0  1.1  1.2   1.4
       1.6   1.3   1.4   1.4    1.4  1.5  1.3    1.5
*The number of SDs varies because some studies do not contain at  least  3  spectra
 required for the computation of a standard deviation.
                                      26

-------
                                   Table VI

         Studies that Contributed Spectra to the Analysis in Table V
Studies that Contributed to Both
   the 542 and 314 Spectra
Study

Borsky

Fishken

Jahn
Ollerhead

Pearson

Pearson

Spiegel

Wells

Wells 3

Wells (

Wells (

Yaniv
Studies that  Contributed to
       205 Spectra




et. al.
id
i and Bennett
; and Wells


10-400 Series
inpublished)
FHV)

Year
1974
1971
1965/66
1964
1971, 1973
1969
1969
1960
1970
1969a
c. 1970
1972
1976
Study
Jahn
L'ubcke, et . al .
Ollerhead
Pearsons and Bennett
Pearsons and Wells
Spiegel
Wells
Wells 300-400 Series
Wells (Unpublished)
Yaniv



Year
1965/66
1964
1971, 1973
1969
1969
1960
1970
1969a
c. 1970
1976



                                      27

-------
                                  Table VII

            Standard Deviations (In Decibels) from Studies Involving
            Mainly Judgments of Annoyance or Unacceptability, for
            Spectra Both with and without Tonal Components, for
            Spectra with Tonal Components, and for Spectra without
            Tonal Components.  (Means were Unweighted.)
                   Number
                     of
                  Spectra
        Number   Frequency Weighting  Calculation Procedure
        of SDs   A    Dl    D2     E    VI   VII   PNL  ZWI
Mean SD (in decibels)
Spectra with
and without
components
260       13    2.5   2.0   2.1   1.9   1.9  1.9  2.1   2.8
Spectra with
tonal components

Spectra without
tonal components
150       12    2.8   2.2   2.2   2.1   2.1  2.1  2.4   2.9
106       11    1.9   1.6   1.8   1.4   1.2  1.3  1.4   2.3
SD of SDs (in decibels)
Spectra with
and without
tonal components

Spectra with
tonal components

Spectra without
tonal components
                1.2   0.8   0.9   0.8   0.8  0.9  0.9   1.5
                1.5   1.1   1.3   1.1   1.0  1.1  1.2   1.6
                1.2   0.8   0.9   0.9   0.7  0.7  0.7   1.8
Studies that Contributed to Both
   the 260 and 150 Spectra
Study                      Year
Borsky
Pearsons and Bennett
Pearsons and Wells
Wells
Wells 300-400
Wells (Unpublished)
       1974
       1969
       1969
       1970
       1969a
       1970
                          Studies that Contributed to
                                 106 Spectra
                       Study                       Year
Pearsons and Bennett
Pearsons and Wells
Wells 300-400
Wells (Unpublished)
1969
1969
1969a
1970
                                      28

-------
Perhaps combining  results  from  diverse  studies  that  used widely  different




methods and  instructions obscures  the possible  effect  of tonal components  on




judged annoyance.  Moreover,  any  interpretation of these findings must  be




limited due  to  the absence  of measurements of "absolute" judged  levels  of




annoyance.





     The  relevance of  the  attribute  judged is further  shown by breaking




Table V's 314 spectra  with  tonal  components  into two groups, those  for




studies in which annoyance  and  noisiness were judged,  and those  in  which




loudness was judged.   Table VIII  shows  that  five of  the eight descriptors  are




more variable for  the  annoyance and  noisiness judgments than for the loudness




judgments; the  other three  are  about the same for both attributes.  However,




the mean SD  for annoyance  across  the eight descriptors' is 2.5 dB compared  to




2.2 dB for loudness.   Such  a  small difference,  0.3 dB, is not meaningful.




     Earlier studies suggested  that  tonal components would be a significant




factor at high  sound pressure levels — in annoyance judgments — but not




at moderate  or  low levels.  If  so, a group of sounds with tonal components




judged with  respect to annoyance  should yield more variable descriptors when




a mixture of both  low  and high  level sounds are  included than when  only low




or only high levels are included.  Of the 233 spectra with tonal components




in Table VIII that were judged  for annoyance and noisiness, 121 were at or




above an overall sound pressure level of 80 dB.   Table IX shows that the SDs




for the 233  spectra are, on the average, larger by 0.1 dB than the  SDs  for





the 121 high level sounds.  This difference is  too small to be meaningful.
                                      29

-------
                                    Table VIII

             Mean Standard Deviations (In Decibels) for Spectra with
      Tonal Components Based on Annoyance, Noisiness, and Loudness Judgments


                Number
Attribute         of    No. of Studies/  Frequency Weighting   Calculation Procedure
 Judged        Spectra  No. of SDs       A    Dl    D2     E    VI   VII   PNL  ZWI


Annoyance         233       8/17        2.9   2.4   2.4   2.3   2.2  2.3   2.5  2.7
and Noisiness

Loudness           81       5/12        2.2   2.5   2.5   2.3   2.0  1.9   2.3  1.6
                                      30

-------
                                  Table IX

            Standard Deviations in Decibels for 233 Spectra with
            Tonal Components at Moderate and High Sound Pressure
            Levels Compared to Standard Deviations in Decibels for
            121 Spectra at or above an Overall Sound Pressure Level
            of 80 dB.
Number
  of      No. of Studies/   Frequency Weighting    Calculation Procedure
Spectra   No. of SDs        A    Dl    D2     E    VI    VII   PNL   ZWI
  233         8/17          2.9  2.4   2.4   2.3   2.2   2.3   2.5   2.7

  121         4/11          3.2  2.3   2.5   2.3   2.1   2.1   2.2   2.1
                                      31

-------
     Closely allied to overall level of the tone is the tone-to-noise ratio.


Only Mark VII and Perceived Noise Level were compared for two ranges of


tone-to-noise ratios.  Over the range of 3 to 13 dB (relative to the third-
            t

octave band level), the mean SD was around 1.6 dB; over the range of 14 to 23


dB, the mean SD increased to around 2.7 dB.  Thus, based on the data examined


in this report, both Mark VII and Perceived Noise Level, and presumably the


other descriptors, may be less accurate in assessing human response to sound


when the tone projects out well above the noise i.e., none of the descriptors


may adequately assess the subjective annoyance produced by relatively strong


tones.


     The effect of the frequency of the tonal components could not be ade-


quately evaluated since in the annoyance studies most of the tones were


between 500 and 2000 Hz.  For 19 spectra with tonal components below 500 Hz,


the mean SD was 0.9 dB for Mark VII and 1.5 dB for Perceived Noise Level.


For 22 spectra with tonal components above 2000 Hz, the SDs increased to 2.9


dB for Mark VII and to 2.4 dB for Perceived Noise Level.  Given the small


sample sizes, this finding is highly tentative although it is consistent with


the analysis of anomalous studies in Appendix B.


     The role of the number of tonal components was also ascertained.


Several of the Wells (1969a, 1970, 1972) studies and the Ollerhead (1971,


1973) study contained sounds with multiple tones as well as with single


tones.   The SDs for Mark VII and Perceived Noise Level were not unusually high


for the group of spectra with both single and multiple tones.  In the


Ollerhead study,  as seen in Table X, the SDs produced by the mixture of


single  and multiple tones is only slightly larger (0.3 dB) than the SDs


produced by spectra with multiple tones only.  These preliminary findings


suggest that the number of components may not affect the variability of the


descriptors.


                                      32

-------
                                   Table X

         Analysis of Standard Deviations in Decibels for Mark VII and Per-
         ceived Noise Level Produced by Data from Ollerhead (1971) Based on
         Spectra that Contained Both Single and Multiple Tones and Spectra
         with Multiple Tones Only.
                          Number of
                           Spectra	Mark VII	PNL
Mean SD (in decibels)
Spectra with Single and
Multiple Tones               60                  3.2                   3.1

Spectra with Multiple
Tones Only                   33                  2.9                   2.8
                                      33

-------
      The  preceding analysis of the effect of tonal components on the variability




of  the descriptors obviously does not lead to clear-cut conclusions.  However,




the main  effect of tonal components on human response appears,  from earlier




studies,  to be an increase in the aversiveness of broadband sounds.  Thus,  it




is  essential to examine the mean differences between calculated and observed




levels.   Only Ollerhead (1971, 1973) provided observed levels based on judgments




of  an evaluative attribute — noisiness; the remaining observed levels are




based on  loudness.  Table XI shows the mean differences from a common group of




studies listed in Table XII that had some sounds with tonal components and some




sounds without.  The differences are just about the same for the two sets of




spectra;  adding tonal components appears to have little effect on the dis-




crepancy  between calculated and observed levels.  Since none of the eight




descriptors makes special provision for tonal components (except, as an integral




part  of the Zwicker procedure), the lack of any effect of tonal components on




the mean  differences suggests that adding tones does not increase the subjective




magnitude.  Moreover, the variability of the mean difference is greater for




spectra without tones than for spectra with tones.   Taken together, the overall




results in Table XI imply that a tone correction procedure may not be needed




when  the  judged attribute is  loudness.




      The  effect of tonal components is different, however, for those sounds that




were  judged with respect to noisiness in the Ollerhead (1971,  1973) study.  Those




mean  differences,  listed separately in Table XI, are more positive for the 44




spectra without tonal components than for the 60 spectra with tonal components.




This  suggests that the observed levels were higher for the spectra with tones




than  for those without.   The  increase in the mean difference is 1.8 dB, averaged
                                      34

-------
                                  Table XI

      Mean Differences in Decibels (Calculated Minus Observed Levels)
      for Studies Containing Some Sounds with Tonal Components and
      Some without.  (Attribute Judged was Loudness Except in the
      Ollerhead (1971, 1973) Study which is also Listed Separately.)
                       Frequency Weighting           Calculation Procedure
                 Number of
                  Spectra  A    Dl      D2     E     VI    VII    PNL   ZWI
Mean of Mean
Differences
Spectra without
tonal components    99   -7.9   -2.0   -2.6   -3.5   3.0   -4.0   4.0   7.5

Spectra with
tonal components   141   -7.7   -1.0   -1.5   -2.9   2.9   -4.6   4.1   7.1
SD of SDs
Spectra without
tonal components          5.6    5.7    5.4    6.1   5.8    6.2   6.6   5.3

Spectra with
tonal components          4.7    4.9    5.0    4.6   4.0    4.0   4.8   3.3
Ollerhead only
(Noisiness judged)

Mean Differences
Spectra without
tonal components    44   -3.1    2.7    1.9    1.5   7.2    1.0   9.9  11.9

Spectra with
tonal components    60   -5.3    1.0   -0.1   -0.1   5.8   -1.3   7.9  10.2
                                      35

-------
                                  Table XII

         Studies that Contributed Spectra to The Analysis  in Table  XI
Studies that Contributed to
        141 Spectra
Study                      Year

Fishken                    1971

John                     1965/66

Lubcke, et. al.             1964

Ollerhead                  1971

Spiegel                    1960

Yaniv                      1976
   Studies that Contributed  to
            99 Spectra
Study                        Year
John

L'ubcke. et. al.
Ollerhead

Spiegel

Yaniv
1965/66

  1964

  1971

  1960

  1976
                                     36

-------
over the eight descriptors.  The most likely interpretation of this result




is that in the Ollerhead  (1971, 1973) study, aircraft sounds with tonal




components were judged the equivalent of 1.8 dB noisier than sounds without




tonal components.  Further, it should be noted here that, in contrast to the




other studies listed in Table XII, the noise stimuli in the Ollerhead (1971,




1973) study had an overall sound pressure level greater than 80 dB.




     In general, the studies examined in this report provide little evidence




for the need for a tone correction.  This finding only appears to contradict




conclusions drawn from some studies cited above.  However, the reasons for the




apparent disagreement may be found in the specific nature of the studies




examined in the present report.  (See Section IV below.)  Furthermore,




Ollerhead's (1971, 1973) data on the aversiveness of sounds with tonal compo-




nents at high levels do suggest a need for a tone correction, but only of the




order of 2 dB.  Despite this generally negative result, the following section




examines and evaluates several tone-correction procedures.







3.  Descriptions of Tone-Correction Procedures




a)  PNLC or FAR 36 Tone Corrections




     A tone correction is contained within the FAR 36 (1969) aircraft certi-




fication regulation.  The tone correction was included to increase, in




accordance with subjective judgments, the measured Perceived Noise Level of




aircraft that produced noise spectra with tonal components.  The Perceived




Noise Level is calculated in the usual way for a given spectrum (Kryter,




1959) -   The FAR 36 procedure then smoothes the spectrum and compares the




original spectrum to the smoothed spectrum in each third-octave band.  If a




band level of the original spectrum exceeds the corresponding band level of




the smoothed spectrum by 3 dB or more, then a correction in decibels is






                                           37

-------
added to the calculated Perceived Noise Level to account for the presence  of




discrete tones.  Thus, within the FAR 36 procedure, the criterion  for  a  tonal




component is that it exceed the noise level in the third-octave band contain-




ing it by 3 dB or more.  The number of decibels added to the calculated




Perceived Noise Level depends on the frequency of the tone and its  level




relative to the smoothed third-octave band noise level.  Tones between 500




and 5000 Hz are penalized twice as much (in decibels) as tones below and




above that frequency range.  The correction cannot exceed 6.67 dB,  which is




the penalty for a tone 20 dB or more above the noise level.  Between tone-to-




noise ratios of 3 dB and 20 dB, the penalty increases linearly with level,




more rapidly in the middle frequency range than elsewhere.  If more than a




single tonal component is identified, only the largest penalty is  added to




Perceived Noise Level; in essence, multiple tonal components are ignored and




a correction is applied only to the strongest tone (taking into account




frequency and tone-to-noise ratio).  This procedure does not take  absolute




level into account, presumably because it was designed explicitly  for  high-




level aircraft noise.  Figure 10 illustrates how the FAR 36 procedure




depends on tone-to-noise ratio and on the frequency of the tone.







b)   Kryter and Pearsons's (1965) Tone-Correction Procedure




     Like the FAR 36 method, the procedure proposed by Kryter and Pearsons




(1965)  is designed for use with Perceived Noise Level.  It is henceforth




referred to in this report as PNLKP.  Instead of first calculating  Perceived




Noise Level and then adding a correction in decibels as in the FAR  36  method




PNLKP first corrects the levels of each third-octave band containing identi-




fied pure tones,  and then calculates Perceived Noise Level according to







                                   38

-------
               I   I      I    I   I    I      I   I   I    I      I    I  -j[—r   |j   |   |
O

c
o

S

t
o
O
o
        I	I
                     500 5000 HZ






I   I   i
I   I    I   I
                                     10              15


                                     Level Difference F,dB
                             20
                                             25
                     Figure 10.   FAR 36 Procedure Tone-Corrections
                                             39

-------
Kryter (1959) on the basis of the revised spectrum.  The result is a tone-




corrected Perceived Noise Level.  In the current application a correc-




tion is made for each band identified as containing a pure tone at third-




octave band center frequencies.  Only tones 3 dB or more above adjacent




third-octave bands have been identified as pure tones in this report although




Kryter and Pearsons (1965) suggested a correction for even smaller tone-to-




noise ratios.  Figure 11 shows that the value of the correction within each




band increases with increasing tone-to-noise ratio up to a maximum ratio of




25 dB.  The value also varies continuously with frequency with a flat maximum




between 3000 and 4000 Hz, depending on tone-to-noise ratio.




c)   Stevens's (1970) Preliminary Tone-Correction Procedure




     In 1970, S. S. Stevens circulated a tentative proposal for a tone-correc-




tion method to be used with his Mark VII or Mark VI computational procedures.




His correction was based on the notion that the underestimation of the calcu-




lated perceived magnitude of a tone-and-noise complex according to Mark VI




or VII arises because the auditory system analyzes components in the complex




as distinct sounds and then, in effect, adds them together to obtain a




total percept.  To develop a procedure that would mimic the auditory system,




Stevens turned to data on the masking of a pure tone by broad-band noise.  He




assumed that the loudness of the partially masked tone would summate with the




loudness of the noise when the two are judged as a composite sound.  Stevens's




procedure takes into account the fact that partial masking depends on the




tone-to-noise ratio as well as on the absolute level of the noise.
                                      40

-------
        •§ 5
        •O Q.
        3-s
A. Octave   T/N
  band    T-N/AN
B. 1/3 octave T/N
  band    T-N/AN
C. 1/10 octave T/N
  band    T-N/AN
25dB
25dB
25dB
25dB
               Figure 11.   Decibel Correction  to  be Added to Sound  Pressure
                            Level of a Band Containing Pure-Tone Component Prior
                            to  Calculation of Perceived-Noise Level.   Parameter
                            is  Band Center Frequency.   Abscissa is Either Ratio,
                            in  decibels, Between Tone  and Noise Measured  Sepa-
                            rately within a Band (T/N)  or the Ratio  Between Level
                            of  Band with Tone and  Noise Together and Level
                            of  Adjacent Bands (T+N/AN)  when Measured with
                            Full-,  1/3-, or 1/10-Oct-Band Filters (from Kryter
                            and Pearsons, 1965).
                                               41

-------
     Stevens did not state what criterion to use  for  identifying  the  presence

of a tonal component.  Thus, the same 3-dB criterion  described  above  for

PNLC was applied to Stevens' correction procedure.  Once identified,  the

tonal component was removed from the spectrum by  averaging the  levels  of  the

immediately surrounding third-octave bands.  The  Perceived Level  is then

calculated by means of Mark VII for the toneless  noise spectrum.  The  decibel

value of the tonal component is read from curves, as  shown in Figure  12.

Although Mark VII was used in constructing the curves that provided the value

of the tonal correction, Stevens' correction can  be applied to  any one of

the descriptors dealt with in the present report.  Once the tonal component

has been removed, the particular frequency weighting  or calculation procedure

is used to compute the predicted level of the toneless spectrum.  Stevens'

tone-correction value in decibels is then added to that computed  level.

     The Stevens correction procedure differs from the FAR 36 and Kryter

and Pearsons (1965) procedures in two main respects.  First, it includes the

level of the band containing the tone as an important determinant of the

value of the tone correction, and second, it omits any dependence of the

correction on the frequency of the tonal component.   The Stevens procedure

also differs in that it is derived from basic psychoacoustic considerations

about the interaction between tone and noise and  in that it includes an

explicit method for handling multiple tonal components.

     The correction for multiple tones assumes that the tones may partly

mask or inhibit one another, the more so the closer they are in frequency.*
^Unless the tones are in the same critical band, in which case they are
 treated like a single tonal component.  Since a critical band is about  as
 wide as a third-octave band, it would require an analysis finer than the
 usual third-octave band analysis to identify such closely spaced tones.
                                      42

-------
                      A)
                      dB
                      10
                      9
                      8
                    0)
                    § -1
                    i 65

                    \ 4
                      3
                      2
                      1
                      0
                                Tone addition
                                   Noise level
                                   in 1/3 octave
                                   5      10     15     20
                                   Level of tone re 1 /3 octave band
                                                      dB
B)
db
 9
                          T
                                  ~T
                                              C)
                      20
                       Level of tone
                       re 1/3 octave
                       band
      20  30   40   50   60  70  80  90  100  dB

                Noise level in 1/3 octave
  db

  20


  18

OJ
B 16
o
o
2 14
CD
i_
£ 12
o
•t-'

- 10
QJ
>
QJ
J  8
                                                20  30  40  50   60   70   80   90  100db

                                                          Noise level in 1/3 octave
   Figure  12.   (A)  Curves for  tone  addition showing the  number of decibels
                (ordinate) added by the tone to the perceived  level of the
                broadband noise  calculated without tone.   The  abscissa shows
                the number of decibels  by which the tone projects above the
                1/3 octave level of th  noise.  The 1/3 octave  level is found
                by averaging the band levels above and below the band that
                contains the tone.  For tone projections greater tha 20 dB
                the tone addition  grows linearly with a slope  of 1.0 (dashed
                lines).  (B)  Same as (A), but with the parameter being the
                level of the tone  as  it projects above the level of the
                noise in the 1/3 octave band.  (C)  Another  alternative
                to (A).  Here the  parameter is the number  of decibels to
                be added to the  calculated level of the noise  for the tone
                projections given  by  the ordinate (From Stevens, unpublished,
                1970.)
                                           43

-------
Mark VII is used in conjunction with the mel  scale  of  pitch  to  compute  the




amount of inhibition from lower to higher frequencies.  Within  each  band,




the calculated  inhibition, expresses  in sones,  is subtracted  from  the




perceived magnitude of a given tonal component,  as  determined by Mark VII.




This value is then converted to Perceived Level  in  decibels which  is used  in




calculating the correction to be applied for  that component.  Finally,  the




corrections computed for each component, after  inhibition  is  taken into




account, are all added to the Perceived Level of the toneless noise.  As




with single tones, this procedure can be applied to any of the  descriptors




examined in the present report.  Accordingly, the Stevens procedure  will be




applied to eight descriptors, with special attention to Mark VII for which




it was primarily intended.




     It must be emphasized that Stevens did not  publish this  tone-correction




procedure, developed in 1969 and 1970, and in all likelihood  intended to




modify it before publication.  Therefore, it  is  to be  considered a tentative




model that may yield insights into just how and when to apply a tone correction.







4.   Other Tone-Correction Procedures




     A tone-correction procedure not evaluated in this report was  proposed by




Wells (1969b) for use with his general annoyance-level (ANL) procedure




for assessing negative effects of noise.  This report  does not  deal  with




his procedure primarily because it has not been as widely discussed  in  the




literature as other descriptors have.
                                      44

-------
     Zwicker's (1958) procedure may be considered a tone-correction procedure




in that  it  is designed to handle  pure tones and combinations of  tones  and




noise, with respect  to loudness.  Only in  so  far as noisiness or aversiveness




differs  from loudness, would his  procedure require a tone correction.  Tables




V and VII do not contradict such  a possibility; Table VIII lends some  support




since, applied to spectra with tonal components, Zwicker's procedure was the




most variable (procedure) when annoyance or noisiness was judged, and  was the




least variable when  loudness was  judged.   Zwicker's procedure handles  tones




on the same basic principles of mutual inhibition that inspired  Stevens's




correction  procedure.




5.   Evaluation of Tone-Correction Procedures




     Similar to the  analysis by spectral shape in Section II, evaluation




of the relative effectiveness of  the three tone-correction procedures  describ-




ed above consists of two parts.   First, the effect of the procedures on




the variability of predicting subjective magnitude is assessed.   Next, their




effect on differences between calculated and judged levels is examined.







a)   Variability




     Table XIII shows the SDs for 260 spectra, some with and some without




tonal components, from six studies and subsets listed in column  1,  Table VII,




in which listeners judged an evaluative attribute (e.g., annoyance, unaccept-




ability).  According to Table XIII, the mean SD for Mark VII corrected by




Stevens's preliminary tone-correction procedure is larger than for Mark VII




uncorrected.  The outcome appears the same when the Stevens correction is




applied to the other seven descriptors, as shown in Appendix C.   Similarly,




the FAR 36 (PNLC) and Kryter and Pearsons (1965) tone-correction procedures
                                     45

-------
          Table XIII.  Effect on Standard Deviations of Three
                       Tone-Correction Procedures.   (SDs  are
                       given for Mark VII with  and without  the
                       preliminary tone-correction procedure of
                       S.S. Stevens.  SDs are also given  for
                       PNL with and without the FAR 36 correction,
                       listed under PNLC, and the proposed
                       correction by Kryter and Pearsons, listed
                       under PNLKP.  Tonal components were  present
                       in all the spectra or only in some.)
 Attribute   Number      Tonal
  Judged   of Spectra  Components
                                                     Calculation Procedures

                                                   Mark VII

                                            Uncorrected  Corrected  PNL  PNLC  PNLKP
                                                1.9
Evaluative              Present    Mean
                                  SD  (dB)
              260         in      	
                                   SD of
   only                  some     SDs  (dB)     0.9
2.2     2.1  2.2    3.2


1.1     0.9   0.9   1.3
Evaluative

and           314

Loudness
                        Present    Mean
                                  SD  (dB)       2.1
                          in      	
                                  SD of
                         all     SDs  (dB)       1.1
3.0     2.4   2.6   3.5


1.4     1.2   1.2   1.8
                                       46

-------
inflate the SDs.  Thus it appears that the tone corrections do not improve




the descriptors' predictability of the negative reactions to noises that




contain tonal components.  If the correction procedures had worked, differ-




ences between noises with tonal components and those without should be reduced,




and the SD of a mix of both kinds of noise should become smaller after




correcting for the presence of tones.  The failure of the three correction




procedures to decrease the SDs may be due to the inclusion of many noises




below 80 dB (although none below 70 dB) where tonal components may be subjec-




tively less important than at high levels.  Such a level effect would be




especially detrimental for Stevens's correction procedure which adds larger




corrections at low than at high noise levels.  Nevertheless, in a separate




analysis, Ollerhead's (1971, 1973) 104 aircraft noises judged for noisiness




were almost all above 90 dB sound pressure level, and yet variability for




those noises also increased from about 3.9 dB to about 4.4 dB when either




Stevens's correction was applied to Mark VII, or the FAR 36 (PNLC) procedure




was applied to Perceived Noise Level.







     Table XIII also shows that the correction procedures increase the vari-




ability for 314 spectra from 13 studies and subsets listed in column 1, Table VI,




all of which contained tonal components.  In some studies, an evaluative attri-




bute was judged, in others loudness was judged.  Mixing the two types of




judgments together may be part of the reason for the increase in variability




when a correction procedure is introduced.  Further, the tone-correction




procedures did not decrease the variability when applied only to tone-noise
                                         47

-------
complexes (a) with multiple tones, (b) with tones at high tone-to-noise




ratios (14 to 23 dB), (c) with tones at lower ratios (3 to  13 dB),  (d) at  or




above an overall sound pressure level of 80 dB, or (e) with tones below  500 Hz,




or above 2000 Hz.  These results are indicated in Table XIV.






b)   Mean Differences Between Calculated and Observed Levels




     If the tone-correction procedures do not reduce variability of the




descriptors, do they at  least reduce the discrepancy between calculated  and




judged levels?  Table XV shows the mean differences between calculated and




observed levels for 141  spectra with identified tonal components, from six




studies and subsets listed in column 1, Table XII.  Observed levels were




calculated according to Mar-k VII, with and without a tone correction, and




according to the Perceived Noise Level (PNL), FAR 36 (PNLC), and Kryter  and




Pearsons (1965, PNLKP) procedures.  If the required 8-dB constant is added to




the Mark VII values to make them 3.4 and 6.7 dB, respectively, then all  three




tone-correction procedures increase the over-estimation of  the measured




level.  More important,  the corrections also increase the SDs of the mean




differences, thus indicating that calculated values vary more around their




means when corrected than when uncorrected.  These 141 spectra included  81




for which loudness judgments were made and for which a tone-correction




procedure may not be needed (see Table XI).  A separate analysis was also




made in Table XI of the other 60 spectra, all from Ollerhead (1971, 1973).




Those data did suggest that a tone correction of about 2 dB may be  necessary




for judged noisiness.
                                   48

-------
     Table XIV.   Effect of Multiple Tones,  Tone-to-Noise Ratio,  Sound Pressure
                 Level of Tone-Noise Complexes Above 80 dB,  and  Tone Frequency
                 on Mean Standard Deviations in Decibels Produced by Three Tone-
                 Correction Procedures.
                                                      Calculation Procedures
Parameter Number of Mark VII
Assessed Spectra Uncorrected Corrected
Evaluation of
Multiple Tones 62 2.7 2.9
Effect of Tone-to-
Noise Ratio
3-13 dB 47 1.6 2.2
14-23 d3 68 2.7 3.5
Effect of Overall
Sound Pressure
Level at or
above 80 dB 121 2.1 2.7
PNL PNLC PNLKP
2.8 3.0 3.4
1.7 1.9 1.6
2.8 2.8 3.0
2.2 2.3 2.7
Effect of Tone
Frequency (Hz)

Tones at or
below 500 Hz
19
0.9
1.4
1.5
1.5
1.7
Tones at or
above 2000 Hz
22
2.9
3.5
2.4
2.4
3.1

-------
    Table XV.  Mean Differences (in Decibels) (Calculated Minus Observed
               Levels for 141 Spectra with Tonal Components.  Attributes
               Judged:  Loudness and Noisiness.)
                            Mark VII                   PNL   PNLC   PNLKP

                    uncorrected    corrected



Means  (dB)            -4.6          -1.3              4.1    7.8     7.8



SD of Means  (dB)      4.0            6.9              4.8    5.8     6.7
                                    50

-------
     The results in Tables XIII, XIV, and XV do not necessarily mean that




the three proposed tone-correction procedures are basically inadequate.




Most of the data used for an assessment of the descriptors, particularly




those used for Table XIV, are based on subjective judgments produced by




spectra from either Wells (1969a, 1970, 1972) or Ollerhead (1971, 1973).




As pointed out in part 2 of this section, as well as in Appendix A, the




inconclusive findings with respect to the need for a tone-correction are most




likely due to the dearth of relevant data.  Before the tone-correction




procedures can be properly assessed, a need for a correction must be clearly




demonstrated.




     Ollerhead's (1971, 1973) study was the only one to provide differences




in level (relative to a specified standard) for a large group of noises




for which an aversive quality,  and not loudness, had been judged.  The




variability of the differences  between calculated and observed levels, like




the combined results in Table XV, did not decrease for Ollerhead's (1971,




1973) noises when the three tone-correction procedures were applied.  The




absolute discrepancies did go down for Mark VII (and also Mark VI) corrected




by the Stevens-correction procedure, but by less than 1 dB, from an average




overestimation of nearly 7 dB to around 6 dB.  A reduction in the overestima-




tion was unexpected since the correction procedure was designed to increase




the calculated values.  Stevens's procedure, however, often results in a




decrease in the calculated level, especially when the tonal components are at




low levels relative to a high-level noise as was the case with Ollerhead's




(1971, 1973) sounds.  On the other hand, PNLC and PNLKP overcorrected and




increased the discrepancy between calculated and measured levels.
                                    51

-------
6.  Summary of Findings Relative to Tonal Components




     The examination of large numbers of spectra with and without tonal




components lends only tentative support to the trends noted in the  literature.




When the judged attribute is either loudness or noisiness, tonal components




did not seem to add to the subjective magnitude of broad-band noise  for




stimuli below 80 dB sound pressure level.  Only when the noise was  at  a high




level (above about 80 dB overall sound pressure level), did the introduction




of tonal components appear to add to the aversiveness of the noise.  Above 80




dB sound pressure level, the increase in noisiness ascribed to the  presence of




tonal components is about 2 dB.  No data seem to be available to adequately




assess the contribution of tonal components to the "absolute" magnitude of




judged annoyance or unacceptability.




     Procedures in use or proposed to correct for the presence of tonal




components did not decrease the variability of Mark VII and Perceived  Noise




Level to which they were applied.  The corrections also did not bring  the




calculated levels closer to the measured levels.  Although a small  correction




may be necessary for the presence of tonal components at high levels,  the




procedures now available cannot be properly assessed until more data demons-




trating the need for a tone correction become available.
                                    52

-------
                   IV. CONCLUSIONS AND RECOMMENDATIONS




     The present report  is a continuation of an earlier report by Scharf, et.




al• (1977).  The present survey sought  (1) to discover whether particular




noise descriptors (sound-level frequency weightings and various calculation




procedures) are more appropriate for certain types of spectral shapes than for




others, and (2) to determine just how important tonal components are in human




response to noise and how best to take  tonal components into account.




     The analysis of data by spectral shape provided a mixed outcome.  Results




revealed little change in the standard  deviations (SDs) of eight descriptors




(frequency weightings A, Dl, D2, and E, and calculation procedures Mark VI,




Mark VII, Perceived Noise Level, and Zwicker's Loudness Computation) when more




than 600 sounds were grouped according  to spectral shape instead of according




to experimental study.  Thus no overall advantage would accrue from regrouping




sets of data across studies on the basis of similar spectral shapes.




The relative efficacy of the eight descriptors in terms of variability was the




same as in Scharf, et. al. (1977) whether the sounds were grouped by spectrum




or by study.  Mark VI, Mark VII, and Zwicker's procedures were the least variable




and the A-weighting was the most variable (C- and B-weightings having been ex-




cluded)—but the difference between the largest SD (2.8 dB for the A-weighting)




and the smallest SD (2.2 dB for Mark VI) was only 0.6 dB.  However,  although varia-




bility was not reduced when considered across all the nine spectral  categories,




it was smaller across the eight descriptors for some categories than for others.




An interaction between descriptor and spectral shape was found to be statistic-




ally significant at  the .001 level.   Despite this significant interaction, the




present data do not  reveal which descriptors are more suited than which others for
                                     53

-------
specific spectral categories.  More judgments of sound annoyance  and  noisiness




are needed, particularly for categories 1 (negative slope), 5  (U-shaped),




7  (low-frequency peaks and valleys), 8 (mid-to-high frequency  peaks and




valleys), and 9 (mixed peaks and valleys) using a known calibrated standard




before this question can be answered.




     Results obtained with a known, calibrated standard would  provide




additional information that permits the computation of mean differences




as well as the standard deviations of the mean differences.  Table IV  showed




that regrouping data by spectral shape rather than by study resulted  in




a  larger reduction in both the SD and range of the mean differences for




the calculation systems than for the frequency-weighting functions.   In




fact, such a regrouping of data enlarged the variability produced by  the




A-weighting.  These results are in line with earlier findings (Scharf,  et. al.




1977) showing that the SDs produced by the frequency-weighting functions,




particularly the A-weighting, strongly depend on level whereas the calcula-




tion systems  are less sensitive to level effects.   Taken together, the




results from this report and Scharf, et.  al. (1977) argue for  the use  of a




calculation procedure such as Mark VI to achieve a significant improvement in




predicting subjective magnitude from physical measurements.  Further,  the




greater flexibility provided by the calculation procedures offers a distinct




advantage  should such factors as tonal components  and duration need to be




incorporated  into these systems.
                                     54

-------
     A detailed analysis of over 500 spectra with and without tonal compo-




nents provided little evidence for the need for a tone correction.  This




outcome would appear to be at variance with previous conclusions  in the




literature.  However, the nature of the studies evaluated was such as to




reduce the likelihood of showing any effect of tonal components.  Many of the




studies required loudness judgments or evaluative judgments at levels below




80 dB.  Even those studies such as Ollerhead's (1971, 1973), which required




evaluative judgments at high levels, stressed noisiness as opposed to annoy-




ance.  Studies by Berglund, et. al.  (1975, 1976) suggest that at high




levels, noisiness and loudness are essentially indistinguishable, whereas




annoyance may remain considerably greater than both noisiness and loudness.




Subjects identify noisiness more as a characteristic of the sound and annoy-




ance more as a description of their own general reaction to noise.  The




presence of tonal components at high levels may affect judgments of annoyance




more than they affect either noisiness or loudness.  However, no measurements




seem to be available of "absolute" magnitude of annoyance caused by sound




with tonal components.  Thus Ollerhead's subjects would probably have given




higher estimates of annoyance, had they been asked, than they did of noisi-




ness when exposed to high-level noise containing tonal components.




     Given the small effect of tonal components in the present group of




studies, the evaluation of three different tone-correction procedures, FAR




36 (1969), Kryter and Pearson's (1965) and Stevens's (1970) could not lead to




definitive conclusions about their relative merits.  Nevertheless, none




of the three improved the effectiveness of the descriptors to which they were




applied; the variability and the discrepancy between calculated and judged




level either remained the same or increased.  This disappointing outcome
                                     55

-------
 should reinforce the realization that data are needed on  a  large  enough set




 of  sounds with and without tonal components to permit adequate  evaluation of




 tone-correction procedures.  Special attention must be  paid to  the  instructions




 given to th,e  subjects.  The present report has tended to  distinguish studies




 on  the basils  of a simple dichotomy, between loudness and  evaluative  judgments




 such as noisiness, unacceptability, and annoyance.  This  dichotomy was  neces-




 sitated by  the nature of the studies investigated which usually mixed together




 a number of adjectives when giving instructions other than  loudness.  The




 reports by  Berglund, et. al. (1975, 1976) suggest that  a  careful  distinction




 should be made among loudness, noisiness, and annoyance in  instructions.   A




 further important point is that most of the studies heretofore  have  used




 psychophysical procedures that emphasise the overall level.  Thus, observers




 are asked to  adjust one sound to be subjectively equal to another sound or to




 report when one sound, presented at various levels, is subjectively  greater




 (or less) than a standard sound.  Such a procedure is usually appropriate for




 investigations of loudness but may inadvertently focus the  subject on loudness




 in a study  that aims to investigate annoyance or even noisiness.  Magnitude




 estimation  has been used successfully for judgments of sound annoyance  by




 Berglund,  _et_. al. (1975,  1976), Bishop (1966), Galanter (1978), Hiramatsu,




 et.  al. (1976), and Scharf and Horton (1978).  By presenting sounds  with  tonal




 components  at different tone-to-noise ratios, frequencies,  and  configurations,




 an experimenter can obtain a fine grain scale of the relative annoyance of




various sounds.  Such experiments would yield the kind of data  needed to  determine




when tonal  corrections are needed and how best to implement  them.
                                      56

-------
                                   APPENDIX A




              CATEGORICAL ANALYSIS ACCORDING TO SPECTRAL TYPE
Introduction
     Categorical analysis involved the identification and classification of




more than 600 spectra that were evaluated in a previous study (Scharf,  Hellman,




and Bauer, 1977) .  The spectra were obtained from 23 studies that encompassed




a wide variety of natural and simulated noises.  In addition to the identifica-




tion and classification of spectra, a statistical analysis of subjective




measurements produced by the noises in each spectral category and across




spectral categories was made based on four frequency weighting functions (A,




Dl, D2, E) and four calculation procedures (Mark VI, Mark VII, Perceived




Noise Level, Zwicker).







Procedure




     The spectra were subdivided into the following nine primary categories:




(1) negative slope, (2) positive slope, (3) broadband flat,  (4)  narrow  band,




(5) U-shaped, (6) inverted U-shaped, (7)  low-frequency peaks,  (8) mid-to-




high-frequency peaks, and (9) mixed peaks.  In order to obtain a finer  analysis




of the data, category 1 (negative slope)  was further divided into two parts,  and




category 4 (narrow band) was divided into three parts.




     Each of the nine categories and subcategories is defined as follows:




     (1)  Negative slope - maximum energy located at low frequencies.




          (A)  Strong negative slope - greater than 5 dB per octave fall-off




               of energy above approximately 500 Hz, but fall-off often




               begins between 100 and 1000 Hz.
                                     A-l

-------
          (B)  Slight negative slope - noise energy  falls  off  from 3 to 5 dB




               per octave above 500 Hz, but fall-off  often begins  between




               100 and 1000 Hz.




      (2)  Positive slope - maximum energy located  at  high  frequencies.   Noise




          energy falls off rapidly below 500 Hz, but  often the  fall-off begins




          at higher frequencies.




      (3)  Broadband flat - spectral distribution of energy remains about the




          same (+5 dB) across a band of frequencies at  least two octaves wide.




      (4)  Narrow band - octave band or narrower.




          (A)  Noise band centered at frequencies  below 500 Hz.




          (fi)  Noise band centered at frequencies  between  500 and  2000  Hz.




          (C)  Noise band centered at frequencies  above 2000 Hz.




      (5)  U-shaped - noise energy reaches a maximum at  low and  at  high  fre-




          quencies, i.e., the noise has a mid-frequency notch.




      (6)  Inverted U-shaped - noise energy falls off  at low and at high




          frequencies, i.e., the noise energy peaks over a broad range  of




          mid-frequencies.




      (7)  Low-frequency peaks - peaks and valleys  in  spectra (+5 dB)  located




          below 500 Hz.




      (8)  Mid-to-high-frequency peaks - peaks and valleys  in spectra (+5 dB)




          located above 500 Hz.




      (9)  Mixed peaks - peaks and valleys in spectra  (+5 dB) located  at




          frequencies both below and above 500 Hz.




     Table A-l shows the distribution of hoises according  to spectral category




and type of noise.   It is evident that the number of  spectra are very unevenly
                                     A-2

-------
TABLE A-l  IDENTIFICATION OF SPECTRA ACCORDING TO SPECTRAL CATEGORY AND TYPE OF NOISE
	 NU
Category Aircraft Industrial
1A -
IB -
2 -
3 -
4A -
4B -
4C -
5 -
6 -
7 -
8 -
9 -
TOTAL
strong, negative 22 18
slight, negative 4
positive
broadband, flat 3
narrow band, centered
below 500 Hz
narrow band, centered
between 500 and 2000 Hz
narrow band, centered
above 2000 Hz
U-shaped
inverted U-shaped 15 14
low-frequency peaks
and valleys 5
mid-to-high frequency
peaks and valleys 32
mixed peaks and valleys 38 16
112 55
nioer ana iype 
-------
divided according to categories.  Categories IB (slight negative slope) and




3 (broadband flat) contain only eight spectra each and category 5 (U-shaped)




contains only six spectra.  On the other hand, category 8 (mid-to-high-frequency




peaks and valleys) has 222 spectra, and category 9 (mixed peaks and valleys)




contains 104 spectra.  Together, categories 8 and 9 contain over half the total




number of spectra.  Table A-l also provides a breakdown by type of sound




(aircraft, industrial, etc.).  The most striking concentration of spectral




shapes is in category 8 (mid-to-high-frequency peaks and valleys) which




contains 56% of the artificial spectra.






Evaluation of Subjective Measurements




     A.  Overall Evaluation




     Within each spectral category noises were grouped according to whether




or not judged loudness levels were provided in the original study.  In those




studies where loudness levels were available,  it was possible to calculate




for each spectral category both mean differences between predicted and




observed loudness levels as well as standard deviations (variability measure).




Whenever loudness levels were not available, only standard deviations computed




from calculated levels could be obtained.  For every category of spectra eight




overall values based on four different frequency-weighting functions (A, Dl,




D2, E) and four different calculation schemes  (Mark VI, Mark VII, Perceived




Noise Level,  Zwicker) were computed.  The eight functions and schemes are




described in greater detail in Table II of Scharf, et.  al. (1977).




     Table A-2 shows the computations of standard deviations determined by




those studies that did not provide calibrated  loudness levels.  A total of




298 spectra from 11 studies listed in Table A-3 contributed to the values
                                     A-4

-------
     TABLE A-2  STANDARD DEVIATIONS (IN DECIBELS)  COMPUTED FROM CALCULATED LEVELS
(Values were weighted within each category according to the number of spectra per study.)
Category
1A
IB
2
3
4A
4B
4C
I 5
6
7
8
9
- strong, neg. slope
- slight, neg. slope
- positive slope
- broadband, flat
- narrow-band, low frequency
- narrow-band, mid frequency
- narrow band, high frequency
- U-shaped
- inverted U-shaped
- low-frequency peaks
and valleys
- mid-to-high frequency
peaks and valleys
- mixed peaks and valleys
MEAN (unweighted) (in decibels)
Frequency Weighting
A Dl D2 E
2.
-
4.
-
3.
2.
2.
-
2.
1.
2.
2.
2.
8 2.7 2.6 2.9
- - -
1 3.3 3.3 3.3
_
6 1.1 2.3 1.7
4 2.2 2.3 1.7
8 3.2 3.0 2.2
- - -
8 2.3 2.5 2.2
5 1.5 1.5 1.6
4 1.8 1.7 1.7
1 2.0 1.8 1.8
7 2.2 2.3 2.1
Calculation Procedure
VI VII PNL ZWI
2.
-
3.
-
1.
2.
1.
-
1.
1.
1.
2.
1.
4

1

2
0
8

8
2
9
0
9
2

3

2
2
1

1
1
1
2
2
.8
-
.3
-
.2
.3
.9
-
.6
.0
.9
.1
.1
2.
-
3.
-
1.
2.
3.
-
2.
1.
2.
1.
2.
7 1.9
-
0 4.0
-
4 2.5
2 3.0
0 5.0
-
0 1.6
3 1.1
0 2.5
9 2.6
2 2.7
Number
of Spectra
27
0
12
0
13
10
12
0
22
53
106
43
298
Number
of SDs
7
0
1
0
1
1
1
0
5
6
6
6
34

-------
          Table A-3   List of Studies that Contributed to Table A-2
No. SDs/study
2
3
1
1
2
2
1
1
1
2
1
Author
Borsky
Kryter
Pearsons and Bennett,
Pearsons and Bennett,
Peasons and Wells
Robinson and Bowsher
Wells (Aircraft)
Wells (Unpublished)
Wells 300
Wells 400
Wells UHV
Year
1974
1959
part 1 1969
part 3 1969
1969
1961
1970
c.1970
1969
1969
1972
No. Spectra/study
10
13
30
20
38
3*
29
30
39
58
25
*Same spectra judged twice, once for loudness and once for annoyance.
                                     A-6

-------
shown in Table A-2.  None of the 11 studies contained spectra for categories




IB (slight negative slope), 3 (broadband flat), and 5 (U-shaped).  Further,




with the exception of the study by Robinson and Bowsher (1961) which provided




both equal-loudness and equal-annoyance •judgments, only equal-annoyance judg-




ments could be obtained from this group of studies.




     Each value in Table A-2 is a weighted mean standard deviation.  That is,




within each category the standard deviations for an individual study were




weighted according to the number of spectra per study.  The criterion estab-




lished for inclusion of a group of sounds was a minimum of three spectra per




study per category.  Moreover, whenever a study consisted of more than one




experiment, standard, or group of sounds, the standard deviation for each




part was determined separately before computing the weighted average for that




particular study.  This added consideration is reflected in the column in




Table A-2 labeled "number of standard deviations".  Therefore, except for




categories 2 (positive slope) and 4 (narrow-band noises), this number is




larger than the number of studies that contributed to the standard deviation




for a given category.




     The method of averaging the results of a particular study may have an




important effect on the outcome.  When a single overall standard deviation




is calculated across diverse portions of a study, such as differences in




procedures, standards, or type of spectra (artificial versus natural sounds),




the standard deviation is inflated.  In addition to parts 1 and 3 of Pearsons




and Bennett (1969) that were kept separate in the analysis undertaken by




Scharf, et.al.(1977), the following studies were also divided into parts:




Borsky (1974), two parts; Kryter (1959), three parts; Pearsons and Wells




(1969), two parts; Wells 400 (1969), two parts.
                                     A-7

-------
     Finally, for each weighting or calculation procedure a mean standard




deviation (unweighted) determined across categories is shown in Table A-2.




The mean standard deviations (unweighted) and standard deviation of standard




deviations across spectral categories and across studies are presented  in




Table A-4.  The values calculated across studies were obtained from those




values in Scharf, et.  al.  (1977) that contributed to the spectral analysis




shown in Table A-2.




     Table A-4 shows that  regrouping the noises according to similar spectral




categories increases the mean standard deviations across weighting and




calculation procedures by an average of about 0.1 dB and decreases the  stan-




dard deviation of standard deviations by an average of 0.2 dB.  Thus, the




overall variability of these data is about the same whether they are grouped




according to study or according to spectral shape.   However, regardless of




how these data are grouped, the A-weighting and Zwicker's scheme produce the




largest SDs and Mark VI produces the smallest SDs.




     More information from Table A-2 can be obtained by evaluating the




results category by category.  The outcome of such  an evaluation is summarized




in Table A-5.  According to Table A-5, of the nine  categories for which




spectra are available, the A-weighting as well as Zwicker's scheme produce




the largest standard deviation for five out of nine categories.  On the




other hand, Mark VI or Mark VII yield the smallest  standard deviations  for




seven of nine categories.   (See Section II, Table III for a discussion  of




the statistical analysis.)
                                     A-8

-------
                                  TABLE  A-4
  Mean Standard Deviations  (in decibels)  and  Standard Deviation  of  Standard
   Deviations  (in decibels)  across  Spectral Categories  and  across Studies
                (for which loudness  levels were not  available)
                          Frequency Weighting             Calculation Procedure
	A    Dl     D2    E	VI   VII   PNL   ZWI   N*
Mean SD
(Unweighted)
1)  across spectral      2.7   2.2    2.3   2.1         1.9   2.1   2.2   2.7   12
    categories

2)  across studies       2.5   2.1    2.1   2.2         1.9   2.0   2.1   2.4   11
SD of SDs

1)  across categories      .78   .75      .60    .62         .58    .64    .63    1.2   12

2)  across studies        1.0    .92      .88    .95         .78    .79    .90    1.4   11



*Number of studies and parts or, number of spectral categories.
                                    A-9

-------
                           TABLE A-5

Analysis of Standard Deviations (in Decibels) According to Categories
                   (Based on Calculated Levels)
Category
1A
2
4A
4B
4C
6
7
8
9
No. Spectra/No SDs
27/7
12/1
13/1
10/1
12/1
22/5
53/6
106/6
43/6
Largest SD
E
A, Zwicker
A
Zwicker
Zwicker
A, D2
A, D2, E
A, Zwicker
Zwicker
Smallest
Mark VI,
Mark VI,
Dl, Mark
SD
Zwicker
PNL
VI
E, Mark VI
Mark VI,
Mark VII,
Mark VII,
D2, E
D2, E
Mark VII
Zwicker
Zwicker


                                   A-10

-------
     Similarly, standard deviations were computed for each of the nine spectral




categories using data from the 10 studies, listed in Table A-6, that provided




loudness levels.




     The results are shown in Table A-7.  A total of 335 noises are distributed




across the nine spectral categories.  More than one-third of the noises are




found in category 8 (mid-to-high-frequency peaks) and the lowest number are




found in category 4A (narrow band low-frequency noises).




     Initially, standard deviations within a specific category were computed




across studies representing a wide range of loudness levels and mean differ-




ences.  No attempt was made to group studies or spectra or to obtain an




actual weighted SD according to the number of spectra per study.  This proce-




dure led to standard deviations as large as 6 dB for categories 1A (strong




negative slope) and 9 (mixed peaks) and as large as 5 dB for categories 4B




(narrow band mid-frequencies) and 8 (mid-to-high-frequencies).  By combining




data across studies and computing a single standard deviation the possible




effect of spectral distribution of energy on standard deviations is obscured




by the very large variation among studies.  A better assessment of variability




within categories is achieved by first calculating the standard deviation for




each individual study or group of individual spectra, averaging these standard




deviations for all studies within a given category and then calculating a




weighted or unweighted mean across spectral categories.  This revised pro-




cedure was used for determining the within category standard deviations shown




in Table A-7-  For the same reasons, i.e., to reduce to a minimum the within




and between study variations, it closely followed the procedure used to




describe the standard deviations indicated in Table A-2.  The initial and
                                     A-ll

-------
TABLE A-6 List of Studies that  Contributed  to Table A-7
 No. Spectra/Study        Study                        Year




        18                Berglund,  et.  al_.            1976




       105                Fishken                      1971




        10                Jahn                         1965/66




         8                Kryter and  Pearsons          1963




        31                Liibcke, et.  al.              1964




        30                Molino                       1976




        37                Quietzsch                    1955




        24                Rademacher                   1959




        39                Spiegel                      1960




        33                Yaniv                        1976
                         A-12

-------
TABLE A-7  STANDARD DEVIATIONS  (in decibels) COMPUTED FROM DIFFERENCES BETWEEN COMPUTED AND OBSERVED LOUDNESS LEVELS
  (Standard deviations were first computed within each study and then weighted within a category according to
   the number of spectra.)
Category
1A
IB
2 -
3 -
4A
4B
> 4C
I—*
w 5 -
6 -
7 -
8 -
9 -
- strong, neg. slope
- slight, neg. slope
positive slope
broadband flat
- narrow-band, low freq. noises
- narrow-band, mid freq. noises
- narrow-band, high freq. noises
U-shaped
-shaped
low-frequency peaks
mid-to-high frequency peaks
mixed peaks
A
3.8
3.7
3.95
3.5
3.4
1.5
1.9
3.7
3.2
1.4
3.0
1.8
We ight ing Scheme
Dl D2 E
3.
2.
4.
3.
3.
1.
2.
4.
3.
1.
4.
2.
5
3
2
2
3
6
2
2
2
2
0
5
3.6
3.4
4.2
3.2
3.3
1.6
2.1
4.1
3.3
1.3
4.0
2.4
3.5
2.4
4.4
3.3
3.4
1.5
2.8
4.0
3.2
1.1
3.7
2.2
Calculation Procedure
VI VII PNL ZWI
3.1
2.4
2.6
2.0
3.8
1.3
2.1
1.7
2.9
1.6
3.2
2.3
2.6
2.2
2.7
2.1
3.7
1.6
1.8
2.0
2.9
1.5
3.4
2.4
4.1
2.6
4.0
3.9
3.5
1.5
1.7
3.5
3.2
2.4
3.7
2.7
3.1
1.7
3.0
2.2
3.6
1.4
1.4
2.5
2.8
1.4
2.7
2.2
No.
Number SDs
37 5
8 2
10 2
8 2
5 1
12 3
6 1
6 2
46 7
20 4
116 12
61 15

Means (Unweighted) (N = 12)
(in decibels)
Means (Weighted according to
number of spectra )
(N = 335)
(in decibels)
2.9
2.8
3.
3.
0
2
3.0
3.25
3.0
3.1
2.4
2.7
2.4
2.75
3.1
3.3
2.3
2.5
Total N=335 noises

Means in Scharf, et . al . Table II,
corrected  re:   Appendix  D,  Table  D-l
(Based on  20 studies)  (in decibels)
3.05   2.65   2.73   2.63
2.26   2.22   2.6
2.

-------
revised procedures were the same for calculating the standard deviations




across categories.




     Each value in Table A-7 for a specific spectral category and weighting




or calculation scheme is a weighted mean standard deviation.  Within each




spectral category individual standard deviations were weighted according to




the number of spectra per study, provided the study had at least three




spectra.  In some studies, spectra fell naturally into groups according to




such variables as signal-to-noise ratio or overall sound pressure level,




e.g., Lubcke, et_. al. (1964), Spiegel (1960), Fishken (1971), and Yaniv (1976).




Standard deviations were then computed for each grouping within a study.




On the other hand, whenever the number of spectra per study fell below the




minimum number of three, the results of more than one study or overall sound




pressure level were combined to produce a single estimate of the standard




deviation.  Hence, as in Table A-2, the numbers in Table A-7 in the column




labeled "number of standard deviations" do not necessarily reflect the number




of studies that contributed to the standard deviations for a given category.




For this analysis, the number of standard deviations is sometimes less than




the number of contributing studies.




     Compared to the initial procedure (i.e. computing a single estimate of




the standard deviation across studies within a spectral category), the revised




procedure (i.e., taking into account standard deviations for individual studies




or groups of spectra within a category before averaging) reduced substantially




the standard deviations computed both within and across categories.   Only for




categories 4A (narrow band low frequency) and 4C (narrow band high frequency)




that are based on one standard deviation do the initial and revised procedures




yield the same result.  Within categories 1A (strong negative slope), 4B
                                    A-14

-------
 (narrow band mid-frequency),  8  (mid-to-high-frequency  peaks),  and  9  (mixed




 peaks), the revised  procedure reduced  the maximum standard  deviation to  4.0  dB




 The mean  standard  deviations  (unweighted) and  standard deviation of  standard




 deviations calculated  across  categories  according to both the  initial  and




 revised within  category  procedures  are  indicated  in Table A-8.  Also shown




 are the mean standard  deviations  (unweighted)  and standard  deviation of  stan-




 dard deviations  calculated  across studies.   Those values were  obtained from




 the standard deviations  in  Scharf,  et.  al.  (1977)  that contributed to  the




 spectral  analysis  shown  in  Table A-7.




     According  to  Table  A-8,  the  revised procedure reduces  the mean  standard




 deviation across categories by  an average of 0.8  dB for the  four frequency-




 weighting procedures and by an  average  of 1.1  dB  for the four  calculation




 schemes.  The mean standard deviation  determined  by the revised procedure is




 about  the same  as  the  mean  SD calculated across studies, but the SD  of SDs




 is about  0.2 dB  smaller.  Further,  regardless  of  how these  data are  grouped,




 the calculation  schemes, with the exception  of PNL, produce mean SDs  about




 0.5 dB smaller  than  the  four  frequency weightings.




     A finer analysis  of Table  A-7  can be accomplished by examining  the




 results,  spectral  category by spectral  category.   The  results  are summarized




 in Table  A-9.   In  contrast  to the results in Table A-5, the A-weighting  fares




much better when loudness levels are provided.  Only for categories  1A (strong




negative  slope) and  IB (slight  negative slope) does the A-weighting  yield




 the largest SDs.   For  category  9, involving  61 spectra with mixed peaks, the




A-weighting produces the smallest SDs.  On  the other hand,  the D1-,  D2-, and




E-weightings produce the largest SDs for 6 out of 12 spectral  categories.




Mark VI,  Mark VII, and Zwicker  calculation procedures  perform  about  equally
                                    A-15

-------
                                       TABLE A-8

          Comparison of Mean Standard Deviations (in Decibels) and Standard
           Deviations of Standard Deviations (in Decibels) Across Spectral
               Categories and Across Studies (Loudness Levels Provided)
                             Frequency Weighting       Calculation Procedure
                             A     Dl    D2    E       VI    VII   PNL   ZWI   N*
Mean SD
(Unweighted)
l)Across Categories
Initial Procedure
Revised Procedure
2)Across Studies
(Based on 10 studies)
3.7 3.9 3.9 3.8 3.5 3.5 4.0 3.6 12
2.9 3.0 3.0 3.0 2.4 2.4 3.1 2.3 12
3.2 3.0 3.1 2.9 2.5 2.4 2.9 2.3 16

SD of SDs
DAcross Categories

Initial Procedure           0.9    1.1   1.0   0.9     1.4   1.5   1.2   1.5   12

Revised Procedure           1.0    1.0   1.0   1.0     0.7   0.7   0.9   0.7   12

2)Across Studies            1.3    1.2   1.0   1.2     0.85  1.1   1.2   1.0   16
(Based on 10 studies)


*N = Number of studies and parts, or number of spectral categories.

LEGEND:

   Initial Procedure:  Within a specific category a single estimate of the
   standard deviation was computed across studies.

   Revised Procedure:  Obtained first the standard deviation for each individual
   study or group of individual spectra and then obtained a weighted mean within
   a category.
                                          A-16

-------
                                Table A-9

Analysis of Standard Deviations (in Decibels) According to Categories
                        (Loudness Levels Provided)
             No. Spectra/
Category
1A
IB
2
3
4A
4B
4C
5
6
7
8
9
No. SDs
37/5
8/2
10/2
8/2
5/1
12/3
6/1
6/2
46/7
20/4
116/12
61/15
Largest SD
A, PNL
A
Dl, D2, E
PNL
Mark VI, Mark VIII
Dl, D2, Mark VII
E
Dl
D2
PNL
Dl, D2
PNL
Smallest SD
Mark VII
Zwicker
Mark VI, Ma
Mark VI, Ma
Zwicker
Dl, D2
Mark VI
Zwicker
Mark VI
Zwicker
Dl, E
Zwicker
A
                                    A-17

-------
well producing the  smallest SDs  for  5 out  of  12  categories.   Mark VI and

Mark VII calculation procedures  perform well  for both  groups  of studies -

those  that provided judged loudness  levels  and those that  did not.   (See

Section II, Table III on  the  spectral categories for a discussion of the

statistical analysis of these data).

     In addition to an analysis  of standard deviations within spectral  cate-

gories, for those 10 studies  that provided  loudness levels  it was also  possible

to perform a within category  analysis of mean differences between calculated

and observed levels.  Table A-10 shows the  results for each of the  eight

frequency-weighting and calculation  procedures and for the  same  335, noises

upon which the standard deviations in Table A-7  are based.  Each value  for

a specific category and weighting or calculation scheme represents  a weighted

mean difference.  Within  a category  the mean differences for  an  individual

study  were weighted according to the number of spectra per  study.   The  mean

differences calculated across categories, the standard deviations of the

means, and the range of mean differences for each weighting and  calculation

scheme are also indicated in Table A-10*.

     Table A-10 suggests  that the A-weighting produces  the  largest  mean

difference, the largest standard deviation, and  the largest range of mean

differences.   The smallest overall mean difference is  produced by Mark  VI

and Perceived Noise Level calculation procedures.  Zwicker's  procedure

produces the smallest standard deviation as well  as the smallest range  of

mean differences.   The differences between Zwicker's procedure and  Mark VI




*Note that,  whereas the means within categories  are weighted  values,  the
 means  calculated across categories are unweighted.
                                     A-18

-------
                    Table A-10




CALCULATED MINUS OBSERVED LOUDNESS LEVEL  (in decibels)
Category
1A
IB
2 -
3 -
4A
4B
4C
5 -
6 -
7 -
8 -
9 -
- strong, neg. slope
- slight, neg. slope
positive slope
broadband flat
- narrow-band, low freq. noises
- narrow-band, mid freq. noises
- narrow-band, high freq. noises
U-shaped
inverted U-shaped
low-frequency peaks
mid to-high frequency peaks
mixed peaks
A
-15.
-18.
-14.
-16.
-11.
-1.
-9.
-16.
-13.
-10.
-4.
-12.
Frequency
Dl
4
1
0
7
5
96
9
4
6
4
8
3
-8.2
-10.5
-5.8
-9.9
-3.2
-0.83
-0.63
-7.95
-8.0
-5.0
+2.1
-5.9
Weighting
D2 E
-10.0
-11.7
-5.4
-10.0
-5.6
-0.56
-0.47
-7.9
-8.3
-6.0
+2.1
-6.2
-9.8
-12.6
-6.8
-11.9
-4.3
-1.46
-2.07
-9.25
-9.8
-6.3
-0.04
-7.8
Calculation
VI VII
-2.5
-6.5
-0.82
-4.0
-2.9
+3.56
-0.27
-2.8
-2.9
-0.8
+5.1
+0.62
-10.1
-12.9
-9.3
-11.4
-10.7
-4.65
-7.97
-10.6
-10.2
-6.2
-1.9
-6.7
Procedure
PNL ZWI
-2.6
-5.1
-2.2
-3.9
-2.1
+2.0
+1.2
-4.0
-2.6
-2.7
+5.7
-0.11
+3.7
-0.15
+4.0
+1.1
-0.02
+6.7
-0.8
+0.1
+2.6
+5.8
+8.0
+5.9
Number
of Spectra
37
8
10
8
5
12
6
6
46
20
116
61

Mean (unweighted)
SD

Range
-12.
4.
16
1
8

-5.3
4.0
12.6
-5.8
4.3
13.7
-6.8
4.1
12.6
-1.2
3.2
11.6
-8.6
3.2
11.1
-1.4
3.1
10.8
+3.1
3.0
8.8
TOTAL
335



-------
and Perceived Noise Level are so small that they are probably not statistically




significant.  Moreover, in contrast to Table IV of Scharf, et. al. (1977) which




suggests that the standard deviation of mean differences across the same eight




frequency-weighting and calculation procedures varies between 4 and 5 dB,




regrouping the data on the basis of spectral categories reduces the standard




deviation of means for the calculation procedures to an average value of 3.1 dB.




This value is about 1.2 dB smaller than the standard deviation of means computed




for the four frequency weightings.  Due to the large differences in number of




spectra that contributed to the weighted mean differences among the nine spectral




categories, a meaningful, statistical analysis of these data could not be




accomplished.  Nevertheless, they do suggest that regrouping the data into




similar spectral categories produces an advantage to the four calculation




procedures but not to the four frequency-weighting functions.




     A more detailed analysis of Table A-10 can be obtained by evaluating the




results category by category, as summarized in Table A-ll.  Table A-ll shows




that the A-weighting consistently underestimates the subjective magnitude of




noise for most categories of spectra, i.e., it produces the largest mean




difference for 10 out of 12 categories.  For the remaining two categories




(4B, narrow band mid-frequency; 8, mid-to-high-frequency peaks), Zwicker's




procedure produces the largest mean difference.  On the other hand, Mark VI




produces the smallest mean difference for six out of the 12 categories.  The




results in Tables A-10 and A-ll, together with the analysis of Tables A-2




and A-7, indicate that the current ANSI standard (1972), Mark VI, is probably




most generally suitable for predicting the loudness or noisiness of noise,




despite the small differences between Mark VI and the other descriptors




evaluated.
                                     A-20

-------
                    Table A-11

Analysis of Mean Differences According to Categories
             (Loudness Levels Provided)
Category
1A
IB
2
3
4A
4B
4C
5
6
7
8
9
No. Spectra/
No . Me an D i f f s .
37/14
8/3
10/4
8/3
5/3
12/8
6/3
6/2
46/7
20/5
116/15
61/20
Largest
Mean Dif f s .
A
A
A
A
A
Zwicker
A
A
A
A
Zwicker
A
Smallest
Mean Diffs.
Mark VI, PNL
Zwicker
Mark VI
Zwicker
Zwicker
Dl, D2
Mark VI
Zwicker
Mark VI, PNL,
Zwicker
Mark VI
Dl, D2, E
Mark VI, PNL
                         A-21

-------
     B.  Effect of Tonal Components on Analysis of Categories 7, 8,  and  9




     The categorical analysis described in Section II also permits a preliminary




assessment of the need for a tone correction procedure to be applied to  the




existing weighting and calculation procedures.  Table A-l indicates  that about




two-thirds of the stimuli evaluated are contained within categories  7, 8,  and




9 (low-frequency, mid-to-high-frequency, and mixed peaks, respectively).   A




more detailed analysis of these spectra was performed to determine (1) what




proportion of spectra in each spectral category included tonal components,




(2) whether within a category those spectra with tonal components produce  a




larger variability and larger mean differences than do those spectra without




tonal components, and (3) whether a specific frequency-weighting or  calculation




procedure was more suited than another for predicting the perceived  magnitude




of noise-tone complexes.   Within each category, noises were grouped  according




to whether or not loudness levels were provided in the original study.




     Table A-12 and A-13 provide two sets of weighted standard deviations  for




categories 7, 8, and 9;  within each category one set of SDs is for spectra




that contained peaks and valleys both with and without tones, and the other




set for such spectra without tones.   Category 9 in Table A-12 includes an




additional set of standard deviations with tones.   The presence of tonal




components was based on criterion developed for the tone-correction  procedures




described in section III.




     Table A-12 and A-13 show that most of the sounds in categories  7 and 8




contained tonal components.   Owing to the large differences in the number of




spectra in the total groups  and the groups without tones, a comparison of




SDs is inappropriate.   (The  larger the n,  the larger the SD tends to become.)
                                     A-22

-------
                                Table A-12

       Standard Deviations  (in  Decibels) Computed  from  Calculated  Levels
         for Categories  7,  8, and  9.   (Loudness Levels  not Measured  in
          Original  Studies.  Values were Weighted  within Each  Category
                 According  to the  Number of  Spectra  Per Study)
	Frequency Weighting	Calculation Procedure
             Number
Category     Spectra	A     Dl    D2    E      VI    VII   PNL    ZWI

   7           53             1.5    1.4    1.5   1.5     1.2   1.1   1.3    1.1
            (with and
            without tones)

                7             1.65   1.1    1.4   1.2     0.6   0.6   0.5    1.1
            (without Tones)
    8          106           2.4    1.8     1.7   1.7     1.9    1.9    2.0    2.5
            (with and
            without tones)

                20           1.7    1.8     1.8   1.6     1.3    1.5    1.8    1.4
            (without tones)
    9           43           2.1   2.0     1.81.8    2.0   2.1   1.9   2.6
            (with and
            without tones)

    *           21           1.3   1.1     1.2  0.85   0.83  0.95  1.0   1.4
            (without tones)

    *           18           3.5   2.6     2.25 2.4    2.4   2.5   2.5   2.7
            (with tones)
*Two spectra were not included in this analysis because they did not satisfy
 the criterion of at least three spectra per study required for computation
 of a standard deviation.
                                    A-23

-------
                                   Table A-13
           Standard Deviations (in Decibels) Computed from Differences
           Between Computed and Observed Loudness Levels for Categories
              7, 8, and 9 (Values were Weighted within Each Category
                   According to the Number of Spectra Per Study)
                                  Frequency Weighting	Calculation Procedure
           Number
Category   Spectra
 A     Dl    D2    E
VI    VII   PNL   ZWI
   7         20
        (with and
         without tones)

             14
        (without: tones)
1.4    1.2   1.3   1.1     1.6   1.5   2.4   1.4
1.3    1.4   1.4   1.2     1.3   1.6   2.4   1.7
   8        116
        (with and
        without tones)
3.0    4.0   4.0   3.7     3.2   3.4   3.7   2.7
             10
        (without tones)
2.7    2.9   2.8   2.5     2.2   2.8   2.5   2.1
             61
         (with and
          without tones)
1.8    2.5   2.4   2.2     2.3   2.4   2.7   2.2
             51
         (without tones)
21     2.8   2.7   2.6     2.5   2.8   2.8   2.4
                                      A-24

-------
Only in category 7 for loudness levels (Table A-13) can a comparison be made,




and there the 14 spectra without tones tended to give slightly larger SDs for




5 of the 8 descriptors than did the 20 spectra with and without tones.  How-




ever, the overall difference of 0.2 dB is not meaningful.  Category 9 provides




a more even distribution for those sounds judged with respect to an evaluative




attribute (Table A-12).  There the SDs are larger by about 1.5 dB for the 18




sounds with tones than for the 21 sounds without.  This finding suggests,




quite tentatively, that in judgments of noisiness, unacceptability, etc., tonal




components may increase the variability of the descriptors for spectra that




contain mixed peaks and valleys.
                                     A-25

-------
                                APPENDIX B




                            "ANOMALOUS" DATA







     The term anomalous data  is used as short hand for the six studies in




Scharf, et. al. (1977) that produced the largest SDs (see Scharf et.  al.,1977,




Table II).  A closer examination of those  studies reveals characteristics




that distinguish them from the average of  the 20 studies and especially from




those studies that yielded the smallest SDs.




     Table B-l shows the standard deviations produced by the six anomalous




studies and those produced by all 20 studies.  For every descriptor,  the




average standard deviation from the six studies is not only larger than from




the entire group of studies, but the disparity is larger by about 0.5 dB for




the six weighting functions than for the five calculation procedures.




     Table B-2 provides a comparison between the six anomalous studies and




the six studies that produced below average standard deviations.   Values are




given for eight descriptors; B, C, and PNLC are omitted.   The mean standard




deviation from the anomalous studies is about 2 dB larger than from the least




variable studies.   The average standard deviation for the weighting functions




is about 0.5 dB larger than that for the calculation procedures.   An  examina-




tion of the less variable studies show that they share the following  charac-




teristics:  a) the spectra tend to be fairly homogeneous; b)  the  stimuli are




exclusively natural, as opposed to artificial,  sounds; c) the range of sound




pressure levels in a study is less than 25 dB;  d) the standard deviations are




based on a single  set of measurements or experimental conditions.




     On the other  hand,  those studies that produced unusually large SDs




differed from the  least variable studies in at  least one  of the characteristics




indicated below.
                                     B-l

-------
                                         Table B-l

                          Standard Deviations (in Decibels) from
                       Six Studies Yielding the Greatest Variability
 Study                        Frequency Weighting	       Calculation Procedure	
                             A    B    C    Dl  D2    E      VI   VII   PNL   PNLC   ZWI

 Fiahken  (1971)
 84/12*                      2.7   2.9  3.0  3.9  3.9  3.6     2.9  2.8   3.8   3.6    2.5
 21/3*                       4.5   4.6  4.6  4.4  4.4  4.4     4.4  5.4   3.4   3.5    3.7

 Pearsons  and  Bennett
 (part 1,  1969)
 30/30                       4.3   4.5  4.7  3.5  3.7  3.3     2.8  2.8   2.9   2.2    3.7

 Pearsons, _et_.^.  (1968)
 108/54*                     6.5   5.1  5.3  2.5  2.8  3.0     2.2  2.2   3.0   2.6    2.1

 Spiegel  (1960)
 20/20                       4.7   6.2  6.8  4.2  4.0  4.2     2.4  1.9   3.2   3.7    2.4
 20/20                       5.3   4.9  5.1  3.5  4.1  3.6     2.6  2.6   2.9   3.2    3.0

 Quietzsch (1955)
 27/27                       4.2   4.4  5.7  4.0  4.3  4.2     3.1  3.2   4.0   4.2    3.3
 10/10                       3.8   6.3  7.0  3.3  2.9  3.8     2.5  2.5   2.6   2.8    2.5

 Wells 300-400 (1969a)
 300- 42/42                  3.7   5.2  6.6  2.4  2.7  2.1     2.1  2.2   2.3   2.4    5.3
 400- 60/60                  2.5   4.2  4.9  2.5  2.0  2.6     2.5  2.6   2.5   1.8    3.1
X 6 Studies
(N = 10)                    4.2   4.8   5.4  3.4  3.5   3.5     2.8  2.8   3.1   3.0    3.2
X 20 Studies
(N = 28)                    3.1   3.6   4.2   2.7   2.7   2.6     2.3  2.2   2.6   2.7    2.4
Scharf, et.al.1977 Table II,
corrected re:  Appendix D,  Table D-l


*  The number in front of the slash  is  the number of conditions (e.g.  different sound
   pressure levels,  instructions,  etc.)

** N = number of standard deviations
                                           B-2

-------
                                   Table B-2

                    Standard Deviations (in Decibels)  from
               Six Studies that Produced the Smallest  Variability
                                Frequency Weighting      Calculation  Procedure
Study	A     Dl    D2    E	VI    VII    PNL    ZWI

Jahn (1965/66)
10/10                           1.3    1.2    1.3    1.2     0.9   0.9    1.0    0.8

Pearsons and Bennett
(1969, part 3)
20/20                           1.7    1.4    1.4    1.7     1.3   1.5    1.3    1.8

Robinson and Bowsher
(1961)
                                1.9    1.4    1.5    1.9     1.2   1.6    1.1    0.9
Wells (1970)
                                1.6    1.2    1.3    0.9     1.2   1.2    1.3    2.2
Wells (Unpublished)
(1970)                          111    1.3    1.3    1.1     0.9   0.9    1.2    1.1

Wells (UHV) (1972)              1.5    1.3    1.5    1.3     1.1   1.0    1.3    0.9

X 6
(N =
studies
6)
1.5 1.3 1.4 1.4 1.1 1.2 1.2 1.3
X 6 anomalous studies
(N = 10)                       4.2   3.4   3.5   3.5     2.8   2.8   3.1   3.2
                                      B-3

-------
Characteristics of Anomalous Studies




     a)   The spectra tend to be heterogeneous.




          Pearsons, et_._ _aL._ (1968, 1969)




          Quietzsch (1955)




          Spiegel (1960)




     b)   The spectra include only artificial sounds.




          Fishken (1971)




          Pearsons and Bennett, Part I, (1969); Pearsons, et. al. (1968, 1969)




          Spiegel (1960)




          Wells 300-400 series (1969a)




     c)   The range of levels is large




          Fishken (1971)




          Quietzsch (1955)




     d)   The standard deviations are based on more than one set of experi-




          mental conditions or measurements.




          Fishken (1971)




          Pearsons, et.  al. (1968, 1969)




     These characteristics suggest under what conditions a group of sounds




is likely to be less well assessed by the descriptors.




     A detailed analysis of five of the six anomalous studies follows.






Spiegel (1960)




     Spiegel's (1960) study illustrates how averaging data produced by




heterogenous spectra inflates standard deviations.  Spiegel studied 20




noises distributed across six spectral categories (2 (positive slope), 4A




(narrow band low-frequency), 4B (narrow band mid-frequency), 4C  (narrow band
                                     B-4

-------
high-frequency), 5  (U-shaped)  and  6  (inverted U-shaped)).  Measurements were




made at two  loudness  levels,  64 phons and 85.5 phons.  The standard deviations




computed  separately for  each  spectral category produce an average value




smaller than the single  standard deviation computed  for all 20 noises, as  in




Table B-l.   Table B-3 presents a re-evaluation of Spiegel's study.  Both sets




of mean standard deviations,  weighted and unweighted, are considerably smaller




than the  overall mean standard deviations shown in Table B-l.




     Recomputing the mean differences between observed and calculated loudness




levels by first calculating the means for a given category and then computing




the overall  means only slightly reduces the overall mean differences.







Wells 300 series (1969a)




     The  measurements by Wells provide another example of the way in which




homogeneous  grouping of  spectra can  reduce SDs.  Wells's 300 series comprised




mainly octave-band  noises both with  and without tones.  The large SDs in




Table B-l  can be ascribed to  heterogeneity of spectra both across and within




categories.




     The  first source of variability can be reduced by computing the SDs




separately for each spectral  category before obtaining the overall average




standard  deviation.  Table B-4 shows that the mean standard deviations





(weighted or unweighted) computed  across categories are smaller by about




0.5 dB than  the previously determined average values.  The A-weighting and




Zwicker's procedure show the  largest reductions.  The A-weighting does least




well for  narrow-band,  low-frequency noises, and Zwicker's procedure is poorest




for narrow-band, high-frequency noises.   Further, with the exceptions of the




A-weighting, Mark VI,  and Mark VII, the SDs tend to be larger for category





4C which  consists of high-frequency noises than for categories 4A and 4B.







                                     B-5

-------
                   Table B-3

Spiegel Study (Standard Deviations  in Decibels)
      Frequency Weighting
      A      Dl    D2      E
Calculation Procedure
VI     VII    PNL    ZWI
CAT.
2
2
4a
4a
4b
4b
4c
4c
5
5
6
6
LL
64
85
64
85
64
85
64
85
64
85
64
N
2
2
2
2
3
3
2
2
3
3
4
85 4
(N=32)

0.31
1.4
0.58
0.84
1.9
2.1
3.7
0.21
4.3
3.1
3.5
6.1

0.51
1.2
0.41
1.0
2.0
2.3
3.8
0.25
4.7
3.6
2.6
4.0

0.47
1.3
0.53
0.88
2.0
2.4
3.9
0.31
4.7
3.5
2.3
5.1

0.46
1.3
0.44
0.97
1.8
2.2
4.0
0.43
4.6
3.3
2.8
3.8

0
1,
0.
0.
1.
1.
3.
0.
2.
0.
2.
3.

.54
.4
,60
,89
4
,6
I
22
6
72
7
7

0.01
1.3
0.70
1.06
1.5
2.0
2.2
0.55
2.5
1.4
1.9
3.7

0.25
1.4
0.25
1.34
1.65
2.1
3.4
0.33
4.5
2.5
2.4
4.1

0.45
1.0
0.20
1.49
1.60
2.2
1.4
2.0
3.4
1.5
2.7
4.0
X, unweighted across categories


64
85
16
16
X, weighted across


X, from


64
85
Table
64
85
16
16
B-l
20
20
2.4
2.3
categ
2.6
2.8

4.7
5.3
2.3
2.1
;ories
2.5
2.4

4.2
3.5
2.3
2.2

2.0
2.1

4.0
4.1
2.4
2.0

2.5
2.3

4.2
3.6
1.
1.

2.
1.

2.
2.
8
4

0
7

4
6
1.5
1.7

1.6
1.9

1.9
2.6
2.1
2.0

2.2
2.3

3.2
2.9
1.6
2.0

1.9
2.3

2.4
3.0
                    B-6

-------
                               Table B-4

           Wells 300 Series (Standard Deviations in Decibels)
                        Frequency Weighting        Calculation Procedure
                        A     Dl    D2    E        VI    VII   PNL   ZWI
N
Category
4a 13 3.6 1.1 2.3 1.7
4b 10 2.4 2.2 2.3 1.7
4c 12 2.8 3.2 3.0 2.2
"X SD
(unweighted) 2.9 2.2 2.5 1.9
X SD
(weighted) 3.0 2.1 2.5 1.9

1.2 2.2 1.4 2.5
2.0 2.3 2.2 3.0
1.8 1.9 3.0 5.0
1.7 2.1 2.2 3.5
1.6 2.1 2.2 3.5
 X
from Table B-l
                  42   3.7    2.4   2.7   2.1      2.1   2.2   2.3   5.3
                                   B-7

-------
     A further reduction in standard deviations can be obtained by subdividing




spectra within a category into two groups, one with tones and another without




tones.  The results of this analysis for an arbitrarily chosen subgroup of 20




of Wells's spectra are presented in Table B-5 together with an analysis of




results for categories 4A and 4C.   The, overall analysis shows that, for seven




out of eight descriptors, spectra with pure tones produce larger standard




deviations than spectra without tones.  Similarly, for categories 4A and 4C




the presence of pure tones enlarges standard deviations for six of the eight




descriptors.







Wells 400 series (1969a)




     The Wells 400 series can be analyzed in the same way as the Wells 300




series.  Wells's 400 series consisted mainly of broadband noises that con-




tained either single or multiple pure tones.  Of the 60 noises,  57 fell into




categories 7, 8, and 9 (low-frequency, mid-to-high-frequency, and mixed peaks,




respectively).  The relatively large SDs in Table B-l can be ascribed to the




presence of multiple and single pure tones as well as to the heterogeneity of




spectra across categories.  It is  possible, for example, to subdivide the




spectra of the Wells'  400 series into two groups, one that consisted of noise-




tone complexes with single tones and another that consisted of noise-tone




complexes with multiple tones.  The results of this analysis for six spectra




that contained multiple pure tones and for 12 arbitrarily chosen spectra that




contained single tones are shown in Table B-6.  With the exception of Zwicker's




procedure, the presence of multiple pure tones produces larger calculated SDs




than the presence of single tones.
                                     B-8

-------
                                Table B-5
      Wells 3001 Series Comparison of SDs (in Decibels) Produced by
                Octave-Band Noises With and Without Tones
                         Frequency Weighting
                         A    Dl     D2    E
Calculation Procedure
VI    VII   PNL   ZWI
N
With Tones 9 5.2 2.9 3.8 2.5
Without Tones 11 3.5 1.6 2.5 0.9
Cat. 4A
With Tones 4 5.5 0.7 3.1 2.4
Without Tones 3 3.4 0.3 1.7 0.9
Cat 4C
With Tones 3 2.8 4.8 5.0 3.5

1.3 2.8 3.2 6.2
2.2 1.8 1.9 3.9
0.24 2.9 0.5 2.8
0.7 0.9 0.7 1.7
1.4 3.4 4.9 7.0
Without Tones
1.6   1.0   1.6   3.3
                                      B-9

-------
                                Table B-6

        Wells 400 Series.  Comparison of SDs (in Decibels) Produced by
           Wide-Band Noises with Single and with Multiple Pure Tones
                         Frequency Weighting        Calculation Procedure
                         A     Dl    D2    E        VI    VII   PNL   ZWI
                  N
Single Tones      12    1.6   1.4    1.2   1.2      1.6   1.6   1.5   2.0


Multiple Tones     6    2.7   3.6    3.0   3.5      2.0   2.3   2.9   1.8
   X
 (Weighted)       18    2.0   2.1    1.8   2.0      1.7   1.8   2.0   1.9
                                  B-10

-------
     The variability across categories can be reduced by computing the stan-




dard deviation for each category separately before computing the overall SD.




Table B-7 shows the results of this analysis as well as the previously calcu-




lated estimate of the standard deviation for the Wells 400 series.  Two features




of Table B-7 are of interest.  First, the calculation of a single estimate of




the standard deviation across diverse portions of a study enlarges slightly




the standard deviation.  Second, as suggested by the Wells 300 series, those




spectra that only contain low-frequency tonal spikes produce smaller standard




deviations than do those spectra that contain tonal spikes above 500 Hz.   The




A-weighting and Zwicker's procedure produce the largest standard deviations




for category 8 (mid-to-high-frequency peaks), and Zwicker's procedure also




produces the largest standard deviation for category 9 (mixed peaks).







Quietzsch (1955)




     Quietzsch's results cannot be analyzed in the same straightforward




manner as those of Spiegel or Wells because his noises varied widely in




spectral shape as well as in amplitude.  The 37 noises varied from 47 to




98 dB overall sound pressure level and 49 to 106 phons loudness level.  Thus,




categorizing the sounds according to spectral shape and computing the standard




deviation separately for each category has only a small effect on the overall




mean standard deviation.  In order to evaluate Quietzsch's results it is




necessary to determine more exactly the effect of sound pressure level




on standard deviations.  However, his results have too few noises at each




level to make this determination.  The specific effect of sound pressure




level on SDs will be demonstrated in the section on Fishken's measurements.
                                     B-ll

-------
                                   Table B-7

               Wells 400 Series (Standard Deviations in Decibels)
                         Frequency Weighting        Calculation Procedure
                         A     Dl    D2    E        VI    VII   PNL   ZWI
                  N
Category
7                 18    1.8    2.2   1.8   2.2      1.4   1.3   1.6    1.3

8                 36    2.8    2.1   1.8   2.3      2.4   2.6   2.0    3.2

9                  3    1.9    2.8   1.8   3.3      3.4   4.2   3.4    4.7
  X   SD
(Unweighted)            2.2   2.4    1.8   2.6      2.4   2.7   2.3   3.1
  X   SD
(Weighted)              2.4   2.2    1.8   2.3      2.1   2.3   1.9   2.7
  X
from Table B-l    60    2.5   2.5    2.0   2.6      2.5   2.6   2.5   3.1
                                   B-12

-------
Pearsons and Bennett (1969)




     Pearsons' and Bennett's data, part 1, produced above average standard




deviations whereas part 3 produced below average standard deviations.  The




range of noise levels in parts 1 and 3 is nearly the same, and the noises




are distributed among the same number of spectral categories.  Parts 1 and




3 appear to differ only in that Part 1 consists exclusively of artificial




noises whereas part 3 consists exclusively of natural noises.




     Table B-8 shows that artificial noises produced larger SDs than did natural




noises.  In addition, consistent with the results of the Wells 300-400 series,




those spectra that contain only low-frequency tonal spikes produced the lowest




SDs.




     Further evidence tha'    -w-frequency tonal spikes produce smaller standard




deviations than mixed low- a^A high-frequency spikes can be obtained from a




within-category analysis of  spectra from Pearsons and Bennett part 1, category




2.  The 12 spectra in this category of noises that produced a positive slope




were divided into two equal  subgroups.  One group consisted of six noises




that contained a low-frequency tonal spike while another group consisted of




six noises that contained both a low- and a high-frequency tonal spike.  The




standard deviations were computed separately for each group.  The results are




indicated in Table B-9.




     Table B-9 shows that,  except for the A-weighting, those noises with low-




frequency spikes produce smaller standard deviations than those noises with




low- and high-frequency spikes.  The largest difference in standard deviation




between the two groups is produced by the Perceived Noise Level procedure.




Table B-9 also shows that the computation of a single standard deviation




acorss diverse portions of a study within a category enlarges the mean stan-




dard deviation by about 1.5 dB.




                                     B-13

-------
                       Table B-8

Pearsons and Bennett (Standard Deviations in Decibels)
             Frequency Weighting        Calculation Procedure
      N      A     Dl    D2    E        VI    VII   PNL   ZWI
Category
6,
6,
7,
7,
9,
9,
part
part
part
part
part
part
1
3
1
3
1
3
7
5
5
2
6
6
3.8
1.2
1.8
1.5
3.5
1.5
2.8
0.9
2.1
0.05
3.4
1.1
3.2
1.0
2.2
0.6
3.5
1.1
2.5
0.7
2.0
0.4
3.3
1.0
1.8
0.9
0.9
0.3
3.1
1.5
1.6
0.8
0.7
0.4
2.9
1.6
2.2
1.1
0.9
0.09
3.5
1.3
1.5
0.9
1.5
0.7
3.1
2.5
                    B-14

-------
                                Table B-9

     Pearsons and Bennett, Part 1 (Standard Deviations in Decibels)
                         Frequency Weighting        Calculation Procedure
                         A     Dl    D2    E        VI    VII   PNL   ZWI
                  N
Category 2

Group I
Low-fre quency
spikes            6     2.2    2.0   1.9   1.8      1.5   1.5   1.6   1.7
Group II
High-frequency
spikes            6     2.1    2.1   2.1   2.1      1.9   2.0   2.2   2.1
    X                   2.15   2.05  2.0   1.95     1.7   1.75  1.9   1.9

Single SD
computed
across category         4.1    3.3   3.3   3.1      3.1   3.3   3.0   4.0
                                  B-15

-------
     The standard deviations across weighting and calculation procedures  for




Pearsons and Bennett, part 1, may be reduced about 1.1 dB by computing the




standard deviation for each of the four spectral categories separately




before computing a mean SD across categories.  This procedure also reduces




the overall SDs produced for Pearsons and Bennett, part 3, but the decrease in




standard deviations is smaller than for part 1.  Nevertheless, the discrepancy




between the standard deviations produced by parts 1 and 3 remains about 1.2 dB,




suggesting that the difference in SDs produced by artificial and natural  noises




is not easily eliminated.







Fishken (1971)




     Fishken measured the overall loudness of broadband noise with tonal  spikes




in two separate series of experiments.  In the first series, the overall  SPL




of the tone and noise was held constant at one of seven overall sound pressure




levels between 30 and 90 dB.  At a given overall sound pressure level, both




the frequency of the tone or tonal complex and the tone-to-noise ratio were




varied.  Four different tones or tonal complexes were combined with three




different tone-to-noise ratios so that a given experimental session consisted




of 12 different tone and noise combinations.  The second series of experiments




by Fishken consisted of three parts.  In each part the tone-to-noise ratio and




the frequency of the tonal complexes were held constant but the overall sound




pressure level of the tone and noise was varied in 10-dB steps over a range




of seven levels between 30 and 90 dB.  The frequency evaluations for tonal




complexes concern those measurements of a pair of 500-Hz and 2000-Hz tones




added to a broadband noise.  The results of both series are evaluated with




respect to each of the following variables:  a) frequency of tone, b) tone-to-noise




ratio,  c)  overall sound pressure level of the tone and noise complex.







                                     B-16

-------
a)   Frequency of Tone




     The analysis of results by Wells  (1969a) and by Pearsons and Bennett




(1969) which were based on annoyance judgments  showed that the presence of




low-frequency tonal spikes produced smaller SDs than the presence of high-




frequency spikes.  A reevaluation of the  first  series of experiments by




Fishken (1971) indicates that  loudness judgments produce a similar outcome.




Two sets of standard deviations were obtained,  one that omitted the 500-Hz




data and another that omitted  the 4000-Hz data.  These results are shown in




Table B-10 together with the standard  deviations previously calculated (see




Table II, Scharf, et. al. 1977 and Appendix D,  Table D-l, this report) for




the entire group of 84 stimuli.  To minimize the possible effect of sound




pressure level on SDs, each value was  obtained by first computing the SD at




each level and then averaging  the results across levels.




     Table B-10 suggests that, unlike  the results of the Wells (1969a) 400




series based on annoyance judgments, the SDs produced by the A-weighting




and Zwicker's procedure do not depend  on sound  frequency when results are




based on loudness judgments.   Five procedures (Dl, D2, E, Mark VI.and Per-




ceived Noise Level), however,  do appear sensitive to a high-frequency spike,




i.e., the SDs are reduced when the 4000-Hz data are omitted suggesting that




the presence of a 4000-Hz tone inflates the SDs.  The opposite occurs for




a tone at 500 Hz.  With the exception  of the A-weighting and Zwicker's




procedure, the SDs produced by the remaining six procedures are larger when




the 500-Hz tone is ommitted.  The results at 500 Hz suggest, in agreement




with the outcome for annoyance, that the presence of a 500-Hz tone decreases




the overall standard deviations.
                                     B-17

-------
                               Table B-10

      Fishken (First Experimental Series).  Effect of Frequency of
               Tone on Standard Deviations (in Decibels)
                         Frequency Weighting        Calculation Procedure
                         A     Dl    D2    E        VI    VII   PNL   ZWI
                 _N	
SD Scharf,
	e_t.al. ,1977
 (Corrected)
  Table II        84    2.7    3.9   3.9   3.6      2.9   2.8   3.8   2.5

SD without
4000-Hz Tone      63    2.8    3.3   3.2   3.0      2.8   2.9   3.5   2.9

SD without
500-Hz Tone       63    2.8    4.4   4.4   4.0      3.2   3.2   4.3   2.7
                                   B-18

-------
     The variability of loudness judgments produced by tone and noise does




not show directly how the presence of a tone may alter the overall judgment




of loudness.  To answer this question, it is necessary to examine mean differ-




ences between predicted and measured loudness levels.  Therefore, mean differ-




ences, computed for the same series of measurements that contributed to Table




B-10, are shown in Table B-ll.




     According to Table B-ll, the mean differences calculated by Zwicker's




procedure are independent of frequency.  Moreover, a tone at 500 Hz heard




together with noise has very little effect on the calculated mean differences,




whereas the removal of a 4000 Hz tone has a more noticeable effect.  When




the 4000 Hz tone is omitted, the mean differences approach zero more closely




for the Dl and D2 frequency-weighting functions and for the Mark VI and




Perceived Noise Level calculation procedures.  Taken together, SDs and




calculated mean differences show that, except for Zwicker's procedure, the




descriptors predict results less well when a 4000 Hz tone is added to broad-




band noise than when a 500 Hz tone is added.




b)   Tone-to-Noise Ratio




     The available evidence (Little, 1961; Pearsons, et. al., 1968) suggests




that, when single and multiple tones are introduced into bands of noise at




tone-to-noise ratios of +15 dB and greater, the sounds become more annoying




than the perceived level predicted by any frequency-weighting or calculation




scheme.  The same 84 stimuli of Fishken (1971) were regrouped to determine




whether this effect of tone-to-noise ratio on annoyance also obtains for




loudness.  Table B-12 shows the effect of tone-to-noise ratio on calculated




SDs and Table B-13 shows the effect on mean differences.
                                     B-19

-------
                                Table B-ll

                    Fishken (First Experimental Series)
              Effect of Frequency of Tone on Mean Differences
          (Calculated Minus Observed Loudness Levels, in Decibels)
                         Frequency Weighting        Calculation Procedure
                         A     Dl    D2    E        VI    VII   PNL   ZWI
                  N
Mean Differences
Scharf, et^.a^. ,
1977 (TabTe~Tv)   84    -4.8  +2.1  +2.0  +0.3     +4.9  -1.9  +5.5   +7.8
Mean Differences
without
4000-Hz Tone      63    -5.2   +0.7  +0.6  -0.9    +4.1  -2.6  +4.2   +7.8
Mean Differences
without
500-Hz Tone       63    -4.6   +2.5  +2.4  +0.5    +5.4  -1.8  +5.9   +7.8
                                   B-20

-------
                                 Table B-12

                     Fishken  (First Experimental Series)
      Effect of Tone-to-Noise Ratio on  Standard Deviations  (in Decibels)
                          Frequency Weighting        Calculation Procedure
                          A     Dl    D2    E        VI    VII   PNL   ZWI
                  N
SDs Scharf,
	et-^1-,1977,
  ("Corrected)      84     2.7     3.9    3.9   3.6      2.9   2.8   3.8   2.5
  Table  II
SDs
  T/N ratios of
  -5and+5dB     56     2.0     2.9    2.8   2.6      2.3   2.0   3.0   1.9
SDs
  T/N ratios of
  + 15 dB          28     2.0     5.5   5.5   4.4      3.7   4.0   5.2   4.0
                                  B-21

-------
                                Table B-13

                    Fishken (First Experimental Series)
             Effect of Tone-to-Noise Ratio on Mean Differences
          (Calculated Minus Observed Loudness Levels, in Decibels)
                         Frequency Weighting        Calculation Procedure
                         A     Dl    D2    E        VI    VII   PNL   ZWI
                  N
Mean Diffs.
Scharf, et.al.,
1977. Table IV    84    -4.8  +2.1  +2.0  +0.3     +4.9  -1.9  +5.5  +7.8
Mean Diffs.
T/N Ratios of
-5 and +5 dB      56    -6.0  +1.3  +1.1   -0.9      +4.0  -2.5  +4.7  +7.8
Mean Diffs.
T/N ratio of
+15 dB            28    -2.1  +3.9  +3.9   +2.5      +6.5   -0.80 +6.9  +7.7
                                  B-22

-------
     The overall SDs are reduced by about 1.0 dB by omitting those stimuli




that produce tone-to-noise ratios of +15 dB.  A tone-to-noise ratio of +15 dB




inflates the SDs for all descriptors exceot the A-weighting.




     In contrast to the data for annoyance, Table B-13 shows that a tone-to-




noise ratio of +15 dB produces an overestimation of perceived loudness level




(not an underestimation).  Only Zwicker's procedure does not overestimate at




+15 dB more than at lower tone-to-noise ratios.




c)   Overall Sound Pressure Level of Tone and Noise Complex




     A striking reduction in standard deviations is obtained by omitting those




data for noises at 30 and 40 dB sound pressure level.  With the exception of




Mark VII and Zwicker's procedure, none of the weighting or calculation proce-




dures was designed to assess loudness below 40 dB sound pressure level.  There-




fore, the use of these procedures for calculating the loudness of noises at




low sound pressure levels is not entirely justified.




     Table B-14 gives the mean SDs calculated on the basis of sounds at all




sound pressure levels from 30 to 90 dB (from Table IV of Scharf, et.  al.,




1977) and on the basis of only those sounds between 50 and 90 dB.  The SDs




for the four frequency weightings go down dramatically from an average of




2.5 dB to 0.21 dB.  The SDs for the four calculation procedures go down from




2.0 dB to 0.9 dB.  Variability also went down quite a bit for Fishken's




second experimental series when the two lowest levels were ommitted.




     The overall sound pressure level of the tone and noise complex also




modifies the difference between predicted and measured loudness levels.




Table B-15 provides an example from the second series of measurements by




Fishken for a constant tone-to-noise ratio of +15 dB.  The discrepancy
                                     B-23

-------
                                Table B-14

       Fishken:  Standard Deviations (in Decibels) with and without
           Low Sound Pressure Levels.  First Experimental Series
                         Frequency Weighting        Calculation Procedure
                         A     Dl    D2    E        VI    VII   PNL   ZWI
                  N
Mean SDs from
Scharf, £t/_aJL- ,
1977 Table IV     84    2.5    2.5   2.5   2.5      2.0   3.3   1.1   1.4
(30 to 90 dB SPI)
SD Means with-
out 30 and 40
dB overall        60     .21    .21   .20   .21     1.1   1.2   0.35  1.0
sound pressure
level
                                   B-24

-------
                            Table B-15

         Mean Differences (in Decibels) from Fishken's Second
          Experimental Series as a Function of Loudness Level
                    (Tone-to-Noise Ratio +15 dB)
              Frequency Weighting	       Calculation Procedure	
	A	Dl     D2     E	VI     VII*   PNL    ZWI
OASPL	

90          +0.5    +7.0   +7.2   +3.4    +7.4    -0.8  +10.4   +8.0

80          -0.9    +5.5   +5.6   +2.0    +5.9    -2.3  +9.1    +8.1

70          -2.0    +4.4   +4.5   +0.9    +5.5    -3.0  +8.0    +8.0

60          -0.3    +6.0   +6.1   +2.6    +8.2    -0.4  +9.5    +10.1

50          +2.2    +8.5   +8.6   +5.2   +11.1   +3.1  +11.6   +12.1

40          +5.5   +11.8  +11.9   +8.4   +14.0   +6.5  +14.1   +14.2

30         +10.3   +16.6  +16.7  +13.2   +17.7   +9.6  +17.8   +17.0
*Note that Mark VII are unadjusted values.
                                   B-25

-------
between predicted and measured loudness levels decreases somewhat from 90




to 70 dB overall sound pressure level and then grows progressively larger




as sound pressure level becomes smaller.
                                    B-26

-------
                                APPENDIX C

         STEVENS'S TONE CORRECTION - 1970 PRELIMINARY PROPOSAL


     Stevens's tone correction may be added  to any of the  descriptors  examined

in this report.  However,  the spectrum must  be smoothed; that  is,  the  tonal

component or components removed, before the  descriptor is  calculated  for  the

noise.  Then the tone correction in decibels calculated according  to  Stevens's

procedure is added to the  descriptor's value.   Since the correction worked

poorly when used with Mark VII for which it  was intended,  it  is  not surprising

that it fares no better with the seven other descriptors as  shown  in Tables C-l

and C-2 for SDs, and in Table C-3 for mean differences.
                                 Table C-l

          Standard Deviations (in Decibels)  for  314  Spectra  from
       13 Studies with Tonal Components Listed in  Column  1,  Table VI.
     (SDs are Given for Each Descriptor with and without  a Correction,
      Based on Preliminary Tonal-Correction  Procedure of  S.S. Stevens,
        Added to the Raw Descriptor Value.   Means  were not Weighted
             According to the Number of Contributing Values.)
                           Frequency Weighting      Calculation Procedure
                           A     Dl     D2    E      VI    VII   PNL   ZWI
Mean SD
Uncorrected
Corrected
SD of SDs
Uncorrected
Corrected
2.6 2.4 2.4 2.3 2.1 2.1 2.4 2.3
3.3 3.0 3.1 3.0 3.0 3.0 3.0 3.0

1.4 1.2 1.3 1.2 1.0 1.1 1.2 1.4
1.8 1.4 1.5 1.4 1.4 1.4 1.4 1.4
                                     C-l

-------
                                  Table C-2

    Standard Deviations (in Decibels) for 260 Spectra from 6 studies with
         and without Tonal Components Listed in Column 1, Table VII.
      (SDs are Given for Each Descriptor with and without a Correction
       Based on Preliminary Tonal-Correction Procedure of S.S. Stevens,
         Added to the Raw Descriptor Value.  Means were not Weighted
              According to the Number of Contributing Values.)
            (Attribute Judged:  Annoyance,  Unacceptability, etc.)
                          Frequency Weighting        Calculation Procedure
                          A     Dl    D2    E        VI    VII   PNL   ZWI
Mean SD
  Uncorrected            2.5   2.0   2.1   1.9      1.9    1.9   2.1   2.8

  Corrected              2.7   2.3   2.4   2.3      2.1    2.2   2.4   2.6
SD of SDs

  Uncorrected            1.2   0.8   0.9   0.8      0.8    0.9   0.9   1.5

  Corrected              1.4   1.1   1.2   1.1      1.0    1.1   1.3   1.5
                                   C-2

-------
                                 Table C-3

     Mean Differences  (in Decibels)  (Calculated Minus Observed Levels)
       Differences are Given  for Levels Calculated Without and With
          a Tonal Correction  Proposed on a Preliminary Basis by
        S.S. Stevens.  Based  Upon  141 Spectra  from 6 Studies with
             Tonal Components Listed in Table  XII, Column 1
                       Frequency weightings
	A     Dl    D2	E	VI     VII    PNL   ZWI
Mean of Mean
Differences	

  Uncorrected        -7.7   -1.0  -1.5   -2.9     2.9   -4.6    4.1   7.1
  Corrected          -5.6     1.6    1.3   -0.2     5.9   -1.3    6.9  10.7
SD of Means

  Uncorrected         4.7   4.9    5.0    4.6     4.0    4.0    4.8   3.3


  Corrected           6.6   7.4    7.6    7.2     7.0    6.9    6.9   6.6
                                     C-3

-------
                               APPENDIX D
             ERRATA AND ADDENDA TO SCHARF, ET. AL. (1977)
1.   Errata
     Several computational errors were noted in four Tables shown in Scharf,
et. al. (1977).  Although these corrections do not change the overall inter-
pretation of results in that report, the revised Tables are included herein.
     Table D-l (Corrected Table II of Scharf, et.  al.):  A computational
error was noted in line 3 based on some of Fishken's data.  This correction
produced a small change in the mean SDs and in the SD of SDs across the 11
descriptors.
     Table D-2 (Corrected Table IV):  Computational errors were noted in
line 1, based on the Berglund, et. al. data, in line 2,  based on some of
Fishken's data, and in line 8, based on the data by Molino.   These correc-
tions produced small changes in the calculation of the mean of the mean
differences and in the SD of the means for the C-  and  D-weightings and  for
Mark VII, PNL, and PNLC.
     Table D-3 (Corrected Table V):   The computational changes made in
Table D-l resulted in small changes  in the values  of the SDs in lines 1,  7,
9, and 13.
     Table D-4 (Corrected Table VI):  The computational  changes made in
Table D-2 resulted in small corrections to the values  of mean differences
in lines 1 and 9.

2.   Addenda
     A repeated-measures analysis of variance (ANOVA), treating studies like
subjects, was performed on the data  in Table II of Scharf, et. al. (1977)
                                     D-l

-------
          Table D-l.  Variability of Calculated Levels  of  Noise by Study
(Standard Deviations in Decibels Computed Either  from the  Calculated Levels of a
Group of Sounds Judged Subjectively Equal or  from the Differences Between Calculated
and Judged Levels.  The Smaller the Standard  Deviation,  the Closer the Scheme Comes
to Predicting the Subjective Equality of a  set of Sounds)

STUDY
Berglund ,
Borsky
Fisken
Jahn
Kryter


It'll-




Kryter and Pearsons
Lubcke, et. al .

Mo lino
Pearsons

Pearsons ,
Pearsons
Quietzsch



NUMBER
N/n OBSERVERS
18/3 30
13/13* 319
84/12* 12
21/3* 8
10/10 28
17/17* 4-100
i 9/9 13-19
11/11 12
20/20 12
30/5* 7
and Bennett 30/30 20

et. al.
and Wells


Rademacher
Robinson
Bowsher
Spiegel

and



Wells (aircraft)
Wells (unpubl.)
Wells 300
Wells 400
Wells UHV
Yaniv








20/20 20
103/54* 20
19/19* 20,20
27/27 20
10/10 20
24/24 20-25

10/5* 558
20/20 10
20/20 10
30/30 35
33/33* 30
42/42 30
60/60 30
25/25 31
11/11 10
11/11 10
11/11 10
Mean SD
Sd of SDs
LEGEND :
N

number of
pressure
conditions (e.g. different
levels, instructions,

A
4.6
3.6
2.7
4 . 5
1.3
2.4
3.5
2.0
2.3
4.4
4.3
1.7
6.5
2.8
4.2
3.8
2.2

1.9
4.7
5.3
1.6
1.1
3.7
2.5
1.5
1.6
2.0
2.6
3.05
1.4
sound


I
4,
3.
2.
4
1.
5,
4.
2
2.
4
4
4
5
3
4
6
2

2.
6.
4.
2.
1.
5,
4,
0
2
1
1
3,
1,



i
.6
.0
.9
. 6
.3
.3
.8
.2
.1
.6
.5
.0
.1
.4
.4
.3
.6

.8
.2
.9
.4
.7
.2
.2
.9
.2
.7
.2
.55
,6




4
2
3
4
1
6
5
2
2
5
4
4
5
3
5
7
3

3
6
5
3
2
6
4
1
4
3
2
4
1



C
.6
.8
.0
. 6
.4
.5
.4
.3
.2
.6
.7
.8
.3
.6
.7
.0
.2

.1
.8
.1
.5
.1
.6
.9
.4
.2
.4
.8
.16
.6



Dl
4.6
3.3
3.9
1.2
3.4
2.8
1.6
2.6
2.9
3.5
1.4
2.5
1.8


4
3
3
1
2
3
1
2
2
3
1
2
1
4.0 4
3.3
1.8

1.4
4.2
3.5
1.2
1.3
2.4
2.5
1.3
2.2
2.4
2.9
2
2

1
4
4
1
1
2
2
1
2
2
3
2.65 2
1.1 1


Mark


D2
.6
.5
.9
.3
.6
.1
.7
.8
.9
.7
.4
.8
.8
.3
.9
.0

.5
.0
.1
.3
.3
.7
.0
.5
.3
.7
.3
.73
.1
VI

MARK
E VI VII PNL PNLC ZWI
4.6 3.8 3.9 5.6 5.6 3.7
3.3 3.0 3.0 3.8 4.2 3.4
3.6 2.9 2.8 3.8 3.6 2.5
1.2 0.9 0.9 1.0 1.5 0.8
3.7 2.5 2.9 2.8 2.6 1.7
2.8 2.1 1.9 2.1 2.2 3.7
1.5 2.5 1.8 1.8 1.4 1.5
2.6 2.1 2.0 2.2 2.3 1.6
2.9 2.4 1.8 2.5 2.6 2.6
3.3 2.8 2.8 2.9 2.2 3.7
1.7 1.3 1.5 1.3 1.3 1.8
3.0 2.2 2.2 3.0 2.6 2.1
1.9 2.4 2.3 2.5 2.7 2.6
4.2 3.1 3.2 4.0 4.2 3.3
3.8 2.5 2.5 2.6 2.8 2.5
1.9 1.6 1.7 1.6 1.7 1.6

1.9 1.2 1.6 1.1 1.4 0.9
4.2 2.4 1.9 3.2 3.7 2.4
3.6 2.6 2.6 2.9 3.2 3.0
0.9 1.2 1.2 1.3 1.7 2.2
1.1 0.9 0.9 1.2 1.6 1.1
2.1 2.1 2.2 2.3 2.4 5.3
2.6 2.5 2.6 2.5 1.8 3.1
1.3 1.1 1.0 1.3 1.4 0.9
1.6 	 2.6 4.6 4.9 2.0
1.7 2.7 1.5 2.7 3.2 0.9
2.1 1.7 1.4 2.7 3.1 1.4
2.63 2.26 2.22 2.60 2.69 2.36
1.1 0.8 1.0 1.1 1.1 1.1
= ANSI S 3.4 (R1972) procedure for the
computation of noise
tone-to-noise ratios)
n

*
number of

standard
different spectra

deviation based on average


of















two or more distinct sets of measurements
A,B,C
Dl

D2

E
standard
sound-level meter weightings
meter weighting adopted by IEC

weighting

weighting







Mark


PNL
PNLC
ZWI

VII






values suggested by K. Kryter

values proposed for trial

and














= based on modification of Mark VI
(S. S. Stevens, JASA, 1972, 51

= perceived noise level
= PNL with tone correction as per FAR 36
= based on Zwicker's loudness calculation
system. Program from E. Paulus and E.
Zwicker, Acustica, 1972, 27. Free-fie
(FF) and diffuse-field (DF) values use.
as appropriate. For earphone listenini
         study by ANSI
                                                          FF values used.
                                        D-2

-------
                                                             Table D-2
                                  MEAN DIFFERENCES (in decibels) (CALCULATED MINUS OBSERVED LEVELS)
o
STUDY
Berglund, et al .
Fishken

Jahn
Kryter and Pearsons
Lubcke, et al

Mo lino
Quietzsch

Raderaacher
Spiegel

Yaniv


N/n
18/3*
84/12*
21/3*
10/10
9/9
11/11
20/20
30/5*
27/27
10/10
24/24
20/20
20/20
11/11
11/11
11/11
See Legend for
A B
-12.9
-4.8
-1.0
-11.9
-8.9
-18.8
-17.3
-6.5
-14.6
-13.0
-8.8
-12.8
-11.9
-7.3
-10.3
-11.8
-4.7
-5.1
-1.7
-10.8
-7.6
-16.9
-15.7
-4.8
-13.0
-9.4
-4.2
-10.9
-10.0
-3.9
-6.9
-8.4
Table II.
C Dl
0.4
-5.1
-1.7
-10.3
-7.0
-16.0
-14.9
-3.1
-11.6
-7.8
-2.4
-10.0
-9.0
-1.3
-4.3
-5.8
-4.1
2.1
5.8
-5.1
-2.3
-13.0
-11.7
-0.8
-8.3
-6.9
-2.1
-6.7
-5.8
-1.7
-4.7
-6.2
D2
-6.7
2.0
6.0
-5.3
-2.2
-13.3
-11.8
-1.0
-8.6
-7.5
-3.0
-7.0
-6.2
-2.4
-5.4
-6.8
E
-6.3
0.3
2.9
-7.7
-3.7
-14.8
-13.3
-2.4
-10.3
-7.9
-3.7
-7.6
-6.8
-3.6
-6.6
-8.1
MARK
VI VII
2.5
4.9
8.8
-0.3
0.3
-8.6
-6.2
6.4
-3.5
-1.4
1.7
-1.5
-2.8
-
0.2
0.9
-6.0
-1.9
1.3
-7.8
-7.3
-13.3
-14.0
-1.0
-11.0
-7.6
-5.3
-9.4
-10.9
-4.3
-4.9
-6.3
PNL
1.9
5.5
9.9
1.1
2.1
-9.4
-5.2
5.1
-2.5
-2.9
3.9
-3.9
-2.3
-1.5
-0.9
-0.9
PNLC
2.6
10.7
15.6
2.3
5.3
-8.2
-3.7
6.3
0.5
-1.0
6.3
-1.8
1.0
-0.1
0.5
0.4
ZWI
8.6
7.8
11.2
5.1
4.1
-2.4
-0.7
12.3
1.8
4.7
8.0
1.0
0.7
6.3
6.5
6.2
          Mean of Mean diffs.




          SD of Means
-10.8    -8.4




  4.53   4.36
-6.9   -4.5    -5.0




 4.85   4.74    4.84
-6.2   -0.13   -6.9   -0.0    2.3    5.1




 4.55   4.54    4.27   4.71   5.66   4.16

-------
                                                         Table D-3
                     EFFECT  ON  STANDARD DEVIATION OF FOUR PARAMETERS (Standard Deviations in Decibels)
U
-P-
See Legend for Table II.
No. of
VARIABLE
1. Attribute Judged
Loudness
Acceptabil ity
2. Type of Noise
Aircraft
Industrial
Vehicle
Household
Art if icial
i; iscel .
3. Tonal Components
Present
Absent
STUDIES/
SDs

9/15
10/12

7/8
3/4
1/1
1/3
7/10
3/4

9/12
10/15
A

3.2
2.9

2.0
2.7
2.2
2.1
4.1
3.5

3.0
3.2
B

3.5
3.7

3.0
2.7
2.6
1.8
4.6
4.1

3.5
3.7
C

4.1
4.3

3.5
2.8
3.2
3.5
5.0
4.9

3.9
4.5
Dl

3.0
2.3

1.9
2.7
1.8
2.5
3.2
2.9

2.5
2.9
D2

3.1
2.3

1.8
2.8
2.0
2.8
3.3
2.9

2.5
3.0
E

2.9
2.3

2.0
2.7
1.9
1.8
3.2
3.1

2.4
2.9
VI

2.5*
2.0

1.5
2.5
1.6
2.2*
2.6
2.3

2.2
2.4*
VII

2.4
2.0

1.6
2.4
1.7
1.8
2.7
2.1

2.3
2.2
PNL

3.0
2.3

1.9
2.9
1.6
3.3
2.9
2.6

2.4
2.8
PNLC

3.2
2.2

2.0
1.7
1.7
3.7
2.8
2.8

2.4
3.0
ZWI

2.2
2.6

1.6
2.1
1.6
1.4
3.2
2.3

2.7
2.2
4. Mode of Sound Presentation
free Field
Diffuse Field
Earphones
11/14
7/8
3/6
2.7
3.7
3.0
3.1
4.7
2.9
3.7
5.3
3.8
2.1
3.1
3.4
2.2
3.1
3.5
2.1
3.3
3.0
1.9
2.3
3.1*
1.9
2.3
2.9
2.1
2.6
3.8
2.1
2.8
4.0
2.3
2.5
2.4
       * 1 SD less

-------
                                                   Table D-4
            EFFECT ON MEAN DIFFERENCES OF TWO PARAMETERS (Calculated minus observed levels in decibels )
o
Ol
See Legend for
VARIABLE
1. Type of Noise
Aircraft
Industrial
Vehicle
Household
Art if ic ial
Miscel .
2. Mode of Stimulus
Free Field
Diffuse Field
Earphones
Table II.
No. of
STUDIES/
MEANS

1/1
3/4
1/1
1/3
3/5
2/3
Presentat
4/5
4/5
3/6

A

-12.8
-15.0
-8.8
-9.8
-7.9
-11.9
ion
-14.3
-10.6
-8.0

B

-10.
-11.
-4.
-6.
-7.
-9.

-12.
-8.
-5.



2
6
2
4
1
1

1
5
1

C

-8.7
-9.6
-2.4
-3.8
-6.6
-7.5

-11.0
-7.4
-3.0

Dl

-5.1
-8.4
-2.1
-4.2
-1.4
-5.3

-8.0
-4.5
-1.5

D2

-5.5
-9.4
-3.0
-4.9
-1.5
-5.7

-8.4
-4.8
-2.2

E

-7.
-10.
-3.
-6.
-3.
-6.

-10.
-5.
-3.



3
5
7
1
0
9

0
7
6

VI

0.9
-3.1
1.7
-0 . 6**
1.9
0.5

-3.4
0.2
3.5

VII

-6.6
-10.2
-5.3
-5.2
-5.6
-6.5

-10.3
-7.2
-3.7

PNL

1.4
-3.6
3.9
-1.1
2.3
-0.1

-2.4
-0.4
2.3

PNLC

_
-3.2
6.3
0,3
5.8
2.0

-0.6
2.0
5.0

ZWI

6.1
2.7
8.0
6.3
9,5
6.3

2.4
4.6
7.8
       ** 2 means

-------
(after being corrected as explained above).  Table II gave the SDs for 28 sets




of spectra for eleven descriptors (six sound-level meter frequency weightings




and five calculation procedures).  The results of the ANOVA are given in Table




D-5.  Although the differences among the mean SDs for the eleven descriptors




were small, they were highly significant, as were the differences among studies




and subsets.  However; the interaction between procedure (descriptor) and study




was not significant.




     To determine which mean SDs differed from each other significantly, a




Duncan's multiple-range test (Lynch and Huntsberger,  1976) was performed on




the matrix of differences between descriptors given in Table D-6.  The number




of asterisks indicates the level of significance.  Generally, differences




greater than 0.45 dB were significantly different at  the .05 level or better.




Thus the A-weighting had significantly larger SDs than four of the five calculation




procedures.  With the exclusion of B- and C-, among the four frequency weight-




ings only A- and Dl-weightings differed significantly.  Except for PNLC, none




of the calculation procedures differed significantly  from one another.  (N.B.




Table D-6 supercedes Table VII in Scharf, et. al. (1977).  Table VII was based




on t-tests and was presented as a preliminary analysis pending an ANOVA and a




more appropriate multiple-range test.
                                     D-6

-------
                 Table D-5
          REPEATED MEASURES ANOVA
(Based on 28 standard deviations from 20 studies.)
Source Variance
Weighting or calculation
procedure
Study
Procedure x Study
Total
Sum
of Squares
95.28
274.04
150.56
530.14
Degrees
of Freedom
10
27
270
307
Mean
9.53
10.14
.56

F P
17.08 «.001
18.20 «.001



-------
                                             Table D-6
               DIFFERENCES  !  IN DECIBELS  BETWEEN MAN STANDARD DEVIATIONS IN TABLE II.

A
B
C
ni
D2
E
VI
a
00 VII
PNL
PNLC
B C Dl D2 E
.50* 1.11*** -.40* -.32 -.42
.61** -.90*** -.82** -.92***
-1.5.*** -1.43*** -1.53***
.03 -.02
Results of Duncan's Multiple Range Test -.10
N=28
blank = Not Significant
* = Significant at .05 or better
** = Significant at .01
*** = significant at .001
-'•Standard deviation for a given calculation schem ie listed in
for the calculation scheme, with which it is paired, listed
Legend :
A, B, C
Dl
D2
E
Mark VI

standard sound-level meter weightings
meter weighting adopted by IEC
weighting values suggested by K Krj'ter
weighting values proposed for trial and
study by ANSI
ANSI S3. 4 (R1972) procedure for the
VI VII PNL PNLC
-.79*** -.83*** -.45* -.36
-1.99*** -1 33*** -.95*** -.gg***
-1.90"** -1.94*** -1.56*** -1.47***
-.39 -.43 -.05 .04
-.47* -.51* -.13 -.04
-.37 -.41 -.03 .06
-.04 .34 .43
.38 .47*
.09

the column of this matrix is subtracted from
in the row. Thus B minus A =.50S Dl minus A

Mark VII based on modification of Mark VI
JASA, 1972, 51)
ZWI
-.69
-1 . 19***
-1.80***
-.29
-.37
-.27
.10
.14
-.24
-.33
the deviation
=-.40, etc.

(S.S. Stevens,
PNL perceived noise level
PNLC PNL with tone correction as per FAR 36
ZWI based on Zwicker's loudness calculation system.
Program from E. Paulus and E. Zwicker, Acustica,
computation of the loudness of noise
1972, 27.  Free-field (FF) and diffuse-field  (DF)
values used as appropriate

-------
                                 REFERENCES

ANSI S 3.4  (R 1972).  USA Standard Procedure  for the computation of loudness
     of noise.

Berglund, B., Berglund, U., and Lindvall, T.  Scaling loudness, noisiness, and
     annoyance of aircraft noise.  J. Acoust. Soc. Am.  1975, 57, 930-934.

Berglund, B., Berglund, U., and Lindvall, T.  Scaling loudness, noisiness and
     annoyance of community noises.  J. Acoust. Soc. Am.  1976, 60, 1119-1125.

Bishop, D.  E.  Judgments of the relative and  absolute acceptability of aircraft
     noise.  J. Acoust. Soc. Am.  1966, 40, 108-122.

Borsky- P.   Annoyance and acceptability judgments of noise produced by three
     types  of aircraft by residents  living near JFK Airport.  NASA report,
     1974,  1-50.

Copeland, W.C.T.  A controlled experiment on  the subjective effects of jet
     engine  noise.  Journal of the Royal Aeronautical Society, 1960, 64,
     33-36.

FAR 36, Federal Aviation Administration, 1969.  Aircraft Certification
     Procedure, Federal Aviation Regulation Part 36.

Fishken, D.  Loudness summation between tones and noise.  Unpublished doctoral
     dissertation.  Northeastern University,  Boston, Massachusetts, 1971.

Galanter, E.  The annoyance of sound and of other life events.  J. Acoust. Soc.
     Am.  1978, ^3_, (SI) S17.

Goulet, P-  and Northwood, T. D.  Subjective rating of broad-band noises
     containing pure tones.  Paper presented  at the 84th Meeting of the Acoustical
     Society of America, Miami Beach, Florida.  December 1, 1972.

Hargest, T. J. and Pinker, R. A.  The influence of added narrow band noises
     and tones on the subjective response to  shaped "white" noise.  Journal of
     the Royal Aeronautical Society, 1967, 71, 428-430.

Hiramatsu, K., Takagi, K., and Yamamoto, T.   The effect of sound duration on
     annoyance.  J. Acoust. Soc. Japan, 1976, 32, 739-750.

International Electrotechnical Commission.  Frequency weighting for the measure-
     ment of aircraft noise (D-weighting).  Document 29C (Central Office) 26,
     1975.

Jahn, M.  Subjektive and objective berwertung von Maschinengerauschen.  Acustica,
     1965/1966, J^, 175-185.

Kryter, K.  Scaling human reactions to the sound from aircraft.  J. Acoust. Soc.
     Am.  1959, 31, 1415-1429.
                                      R-l

-------
Kryter, K. D.  The Effects of Noise on Man.  Academic  Press,  Inc.:   New York,
      1970.

Kryter, K. D., and Pearsons, K. S.  Some  effects  of  spectral  content and
      duration on perceived noise  level.   J. Acoust.  Soc.  Am.,  1963S  35,  866-883.

Kryter, K. D., and Pearsons, K. S.  Judged noisiness of  a band  of random noise
      containing an audible pure tone.  J. Acoust.  Soc. Am.,  1965,  38,  106-112.

Little, J. W.  Human response to  jet engine noises.  Noise Control,  1961,  _7_,
      11-13.

Little, J. W. and Mabry; J. E.  Empirical comparisons  of  calculation procedures
      for  estimating annoyance of  jet aircraft  flyovers.   J. Sound  Vib.,  1969,
      _10,  59-68.

Lubcke, E., Mittag, G. , and Port, E.  Subjective  und objective  Bewertung von
      Maschinengerauschen.  Acustica, 1964, 14,  105-114.

Lynch, M.  D., and Huntsberger, D. V.  Elements  of Statistical Inference  for
      Education and Psychology.  Allyn and Bacon,  Inc.:  Boston,  1976.

Molino, J. National Bureau of Standards.  Unpublished  Study,  1976.

Niese, H.  Beitrag zur Relation zwischen Lautstarke und Lastigkeit von
      Gerauschen.  Acustica, 1965, 15, 236-243.

Ollerhead, J. B.  An evaluation of methods for  scaling aircraft  noise  percep-
      tion.  NASA CR-1883, 1971.

Ollerhead, J. B.  Scaling aircraft noise perception.   J.  Sound  Vib.,  1973,
     _26_(3), 361-388.                                  	

Paulus, E. and Zwicker, E.   Program zur automatischen  Bestimmung der
      Lautheit aus Terzpegeln oder Frequenzgruppenpegeln.   Acustica,  1972,  27,
      253-266.                                                              	

Pearsons, K.  S.   Assessment of the validity of  pure tone  corrections  to
      perceived noise level.  Conference on progress of NASA research  relating
      to noise alleviation of large subsonic aircraft,  NASA SP-189, 1968.

Pearsons, K.  S.  and Bennett,  R.  L.  The effects of temporal and  spectral
      combinations on the judged noisiness of aircraft  sounds.   FAA report.
      1969.

Pearsons,  K.  S.  and Bennett,  R.  L.  Effects of  temporal and spectral  combina-
     tions on the judged noisiness of aircraft  sounds.   J. Acoust. Soc. Am
      1971, 49, 1076-1082.                                           "	~'
                                      R-2

-------
Pearsons, K. S. Bishop, D. E., and Horonjeff, R. D.  Judged noisiness of
     modulated and multiple tones in broad-band noise.  J. Acoust. Soc. Am.,
     1969, 45, 742-750.  Also NASA report, 1968.

Pearsons, K. S. and Wells, R. J.  Comparison of objective measures of
     noisiness for pure tone and varying bandwidth noise.  Paper presented at
     the 6th International Congress on Acoustics, Tokyo, Japan, August 21-28,
     1968.

Pearsons, K. S. and Wells, J. R.  Judged noisiness of sounds containing
     multiple pure tones.  Paper presented at the 78th Meeting of the
     Acoustical Society of America, San Diego, California, November 4-7,
     1969.

Quietzsch, G.  Objektive und subjektive Lautstarkemessungen.  Acustica, 1955,
     Akust. Beih. _1, 49-66.

Rademacher, H.  Die Lautstarke von Kraftfahrzeuggerauschen.  Acustica, 1959,
     9^, 91-108.

Robinson, D. W. and Bowsher, J. M.  A subjective experiment with helicopter
     noises.  J. of the Royal Aeronautical Society, 1961, 65, 635-637.

Rule, S. J.  Effect of instructional set on responses to complex sounds.
     J. Exp. Psych.  1964, 67, 215-220.

Rule, S. J. and Little, J. W.  The design and construction of an annoyance
     scale for jet engine noise.  Technical report.  Commercial Airplane
     Division, The Boeing Company, Seattle, Washington, January 23, 1963.

Scharf, B., Hellman, R. and Bauer, J.  Comparison of various methods for
     predicting the loudness and acceptability of noise.  U.S. Environmental
     Protection Agency Report.  550/9-77-101, August 1977.

Scharf, B. and Horton, T. J.  Scaling loudness and annoyance as a function
     of duration.  J. Acoust. Soc. Am.  1978, 63_ (SI) S16.

Spiegel, M.  Prufung verschiedener Lautstarkeberechnungsmethoden bei
     diffuser Beschallung.  Unpublished, 1960.

Stevens, S. S.  Procedure for calculating loudness:  Mark VI, J. Acoust. Soc.
     Am.  1961, J33, 1577-1585.

Stevens, S. S.  Perceived level of noise plus tones.  Unpublished report for
    the Laboratory of Psychophysics, Harvard University, 1970.

Stevens, S. S.  Perceived level of noise by Mark VII and decibels (E).  J_._
     Acoust. Soc. Am.. 1972, 51, 575-601.
                                      R-3

-------
Wells, R. J.  Recent research relative to perceived noise level.  Paper pre-
     sented at the 74th Meeting of the Acoustical Society of America, Miami
     Beach, Florida, November 14-17, 1967.

Wells, R. J.  A new method for computing the annoyance of steady state noise
     versus perceived noise level and other subjective measures.  Paper
     presented at the 77th Meeting of the Acoustical Society of America,
     Philadelphia, Pennsylvania, April 8-11, 1969a.

Wells, R. J.  Calculation of annoyance level for sounds containing multiple
     pure tones.  Paper presented at the 78th meeting of the Acoustical
     Society of America, San Diego, California, November 4-7, 1969b.

Weils, R. J.  Jury ratings of complex aircraft noise spectra versus calculated
     ratings.  Paper presented at the 80th Meeting of the Acoustical Society
     of America, Houston, Texas, November 3-6, 1970.

'Jr.lls, R. J.  Unpublished (about 1970).

Wells, R. J.  A subjective study of ultra high voltage transmission line
     noise.  Paper presented at the Second Arden House Workshop on Noise
     Control Engineering, January, 1972.

Winer, B. J.  Statistical Principles in Experimental Design, McGraw-Hill:
     New York, 1962.

Yaniv, S.  Equal loudness contours for household appliances. National Bureau
     of Standards. Unpublished Study, 1976.

Zwicker,  E.  Uber psychologische und methodische Grundlagen der Lautheit.
     Acustica, 1958, 8 (Beiheft 1), 237-258.
                                      R-4

-------
                                  TECHNICAL REPORT DATA
                           (Please read Instructions on the reverse before completing)
 REPORT NO.

   EPA 550/Q-7q-in?
                                                          3. RECIPIENT'S ACCESSION NO.
 TITLE AND SUBTITLE
 Part II:  Comparison  of  Various Methods for Predicting
 the Loudness and Effects of Spectral  Pattern Accept-
 ability of Noise and  Tonal  Components	
                                                          5. REPORT DATE
                                                            Mov  1979
                             6. PERFORMING ORGANIZATION CODE
 AUTHOR(S)

 B.  Scharf and R. Hellman
                                                          8. PERFORMING ORGANIZATION REPORT NO.
 PERFORMING ORGANIZATION NAME AND ADDRESS

 Auditory Perception  Laboratory
 Northeastern University
 Boston, Massachusetts  02115
                                                          10. PROGRAM ELEMENT NO.
                              11. CONTRACT/GRANT NO.
 2. SPONSORING AGENCY NAME AND ADDRESS

 U.S.  Environmental Protection  Agency
 Office of Noise Abatement  and  Control
 Washington, D.C.   20460
                                                          13. TYPE OF REPORT AND PERIOD COVERED
             (ANR-471)
                              14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
      The present report  is  a  continuation of an earlier report by Scharf,  Hellman
  and Bauer (1977).  The  objectives are (1) to determine whether subjective judgments
  of particular types of  noise,  categorized by spectral shape, are better approxi-
  mated by some descriptors  (frequency weightings and calculation procedures)  than
  by others, and (2) to investigate the role of tonal components in  these studies and
  to assess the adequacy  of  several tone-correction procedures.  The analysis  of data
  by spectral shape  produced a mixed outcome.  Results showed that no  overall  advan-
  tage would accrue  from  regrouping sets of data across studies on the basis of  similar
  spectral shapes.   However, although variability was not reduced when considered acros
  nine spectral categories,  the interaction between spectral shape and descriptor was
  highly significant (p < .001).   The examination of over 500 spectra  with  and without
  tonal components provided  only tentative support for the trends noted  in  the litera-
  ture.  When the judged  attribute is either loudness or noisiness,  tonal components
  do not seem to add to the  subjective magnitude of broad-band noise below  80  dB sound
  pressure level.  At higher levels, according to one large-scale study, tonal compon-
  ents seemed to add the  equivalent of 2 dB to the judged noisiness.   No data  could  be
  located that would permit  adequate assessment of the contribution  of tonal components
  to the "absolute"  magnitude	(continued on page 2, attached)
                                                                                         \
17.
                               KEY WORDS AND DOCUMENT ANALYSIS
                  DESCRIPTORS
                                              b.IDENTIFIERS/OPEN ENDED TERMS
                                                                          COSATl Field/Group
  Loudness, Spectral Patterns,  Tonal
  Components, Calculation  Procedures,
  Frequency Weightings
18. DISTRIBUTION STATEMENT
  Available at EPA/ONAC and
  gle Park,  North Carolina
Research Trian-
19. SECURITY CLASS (ThisReport)
  Unclassified
                 20. SECURITY CLASS (This page)
                   Unclassified
                                            22. PRICE
EPA Form 2220-1 (9-73)

-------
                                      -  2  -

...of judged annoyance or unacceptability  (as  distinct  from  noisiness  or  loudness).
Given the small  effect of tonal  components in  the  present  group  of  studies,  the
evaluation of three different tone-correction  procedures (FAR  36, 1969; Kryter and
Pearson's, 1965; and Steven's,  1970)  could not lead  to  definitive conclusions about
their relative merits.  Although a  small correction  may be necessary for  the pre-
sence of tonal components at high levels,  the  tone-correction  procedures  now avail-
able cannot be properly evaluated until  more appropriate data  that  demonstrate the
need for a tone  correction are  obtained.

-------