19 th INTERNATIONAL CONGRESS ON ACOUSTICS MADRID, 2-7 SEPTEMBER 2007 A SEMANTIC DIFFERENTIAL STUDY OF LOW AMPLITUDE SUPERSONIC AIRCRAFT NOISE AND OTHER TRANSIENT SOUNDS PACS: 43.28.Mw Marshall, Andrew 1 ; Davies, Patricia 1 1 Ray W. Herrick Laboratories, Mechanical Engineering Department, Purdue University; 140 S. Intramural Dr., West Lafayette, IN, 47906, USA; daviesp@purdue.edu ABSTRACT Aircraft manufacturers have proposed building supersonic business jets that create low amplitude sonic booms that may be less objectionable to the public than those that led to the ban of commercial supersonic flight over the US. Not only do these signatures differ from traditional sonic booms in amplitude, they also differ in shape. The question arises: what type of supersonic aircraft signatures would be acceptable to the general public? Two experiments were conducted. In each experiment, low amplitude booms and other transient sounds, such as car door slams and distant gunfire, were used as stimuli. In the first test, subjects described 64 sounds and gave each an acceptability score. The words subjects provided were used to develop semantic differential pairs that were used in the second test where subjects rated 24 sounds. The objective of this test was to determine the number of independent attributes people perceive and how these influence acceptability or annoyance. As in previous tests with higher amplitude stimuli, subjects responded to attributes related to the frequency content and the temporal properties of the sounds as well as to loudness. The relationship between the measured perceptual attribute strengths and statistics of time-varying loudness was also examined. INTRODUCTION Recently, manufacturers have proposed building supersonic business jets that create only low amplitude sonic booms. These booms may be less objectionable to the public than those that led to the current ban of commercial supersonic flight of the United States. These shaped sonic booms, however, may differ from traditional sonic booms in both amplitude and shape and it is of interest whether the types of metrics used to predict the response to those traditional sonic booms (Leatherwood, Shepherd & Sullivan, 1991) are appropriate metrics for these low amplitude shaped sonic booms. Sullivan (2004) used recordings of low amplitude booms and simulated shaped booms in a subjective test and examined correlation of responses to metrics quantifying level of the sounds, but there has been little investigation on whether other sound characteristics play a role in annoyance. Also, recent developments in models of perception of time-varying loudness (Moore & Glasberg, 2002) may result in metrics that quantify perception of the loudness of transient sounds more precisely. With these points in mind, experiments were conducted to determine the perceptual dimensionality of these transient sounds and examine which metrics can be used to predict annoyance. Currently, it is not feasible to conduct community surveys to assess the response to these modified sonic boom signatures, so tests must be conducted in a laboratory setting. The tests described here and in a companion paper (Marshall and Davies, 2007) were conducted in a sound proof booth over high quality in-ear phones. It will be interesting to compare these findings with those from tests conducted in sonic boom simulators where sounds are both heard and felt, and with results of field tests, when they become feasible. It is interesting that the set of words collected to develop the scales used in the semantic differential test described in this paper were very similar to those collected in a study of the realism of existing sonic boom simulators (Marshall, Davies, Sullivan & Pilon, 2006).
First a test was conducted to generate a lexicon to describe low amplitude booms and other transient environmental sounds. The next experiment was a semantic differential experiment which used the words from the lexicon experiment to populate the word scales. The focus of this paper is the correlation between metrics and the responses in the semantic differential test, as well as an examination of how many underlying signal characteristics (factors) are being perceived by subjects. In the companion paper (Marshall & Davies, 2007), a more detailed description of the vocabulary and semantic differential scale generation will be given, along with a more in-depth discussion of the perceived attributes and how they may affect annoyance and acceptance. SEMANTIC DIFFERENTIAL TEST PROCEDURE Institution Review Board approval was recieved for the test (IRB: 0603003533). After a brief description of the test, subjects completed a consent form and then the subject s hearing was tested. The subject was then placed in the sound booth, and the test was explained to them. The subjects then heard a subset of the stimuli used in the test to familiarize themselves with these types of sounds that they would hear. The subjects were asked to rate each sound they heard using the scales provided and they should base their judgments on how they would feel about the sounds if they were outdoors in a park or garden and heard the sounds intermittently throughout the day. A sound was repeated at random intervals while the subjects completed the ratings. Once a set of ratings was completed, the subject would move on to the next sound. The sounds were repeated at random intervals to prevent the subjects from habituating to the sounds and prevent the suppression of startle effects. A similar method was also used by Plotkin and Bradley (1991) in an aircraft noise study. After listening to the instructions, the subject completed a practice session where they rated four sounds. Upon completing the practice session, the subject completed the ratings for the set of 24 sounds. The order of the words at the end points of the scales were randomized, the orderings of the scales on the page and the sounds were randomized, a different random order for each subject. Clearly acceptability and annoying would be functions of the frequency of sound event occurrence. The greater the frequency of events, especially with loud transients, the less acceptable and more annoying the sound events would be. As no plans have been made to fly low-boom craft as of yet, there is no adequate idea of the frequency of occurrence of these sound events. Thus, not wanting to overly influence subjects, the vague phrase intermittently through the day was used. SCALES, SIGNALS AND SUBJECTS Word Scales The 20 words scales are shown in Table 1. These were presented on one sheet of paper at end points of lines. There were five equi-spaced marks on the lines and subjects could make a mark at any point on the line. The position of the marks on the scales was measured to within 1/64 th of the entire scale length, and were in the range -16 to +16. Table 1. The twenty word scales used in the semantic differential test. 'Loud' 'Soft' 'Acceptable' 'Not Acceptable' 'Sharp' 'Dull' 'Annoying' 'Not Annoying' 'High pitch' 'Low pitch' 'Startling' 'Calming' 'Peaked' 'Rounded' 'Pleasant' 'Unpleasant' 'Harsh/Rough' 'Smooth' 'Soothing' 'Disturbing' 'Fast' 'Slow' 'Distracting' 'Easily ignored' 'Close' 'Distant' 'Long' 'Short' 'Agitated' 'Tranquil' 'Lots of echoes' 'No echoes' 'Industrial' 'Natural' 'Rumbling' 'Clean' 'Violent' 'Peaceful' 'Deep' 'Shallow' 2
Subjects The results presented here are from data provided by 14 naïve subjects who participated in this study. The subjects were volunteers from the area surrounding Purdue University. Subject ages ranged from 20-40 years, 11 were male and 3 were females. 13 were US born and raised, and the other was from a Southeast Asia. Subjects had less than 30 db of hearing loss at frequencies from 500 to 8000 Hz. Subjects were paid $10. Test Sounds Twenty-four sounds were used in this experiment. These included 12 low amplitude N-wave recordings, 1 gunfire recording and 1 door slam that were used in the lexicon development test in which subjects also rated the acceptability of the sound (same context described as in this test). These sounds were selected to include the entire range of acceptability fractions (.1 -.9) obtained in that test. Also, we tried to pick the set of sounds that yielded the smallest correlations between sound quality metrics. Another 5 sounds were selected from modified candidate low boom waveforms. In addition there was 1 recording of thunder and 2 recordings of a sonic boom and a door slam with most of its low frequency energy removed (frequencies of less than 100 Hz). The 4 remaining sounds were replicates of sounds from the previous lexicon experiment. To allow for the largest range of overpressures (which were limited by the low frequency playback capability of the earphones), all of the sonic booms were high-passed filtered (Butterworth, 4 th order digital filter) with a cut-off frequency of 25 Hz. This frequency was selected so that subjects were unable to notice the effects of the filtering when heard over the in-ear phones. The maximum overpressure was around 25 Pa. The sounds were played binaurally to subjects over Etymotics ER-2 earphones in an IAC double-walled sound booth. RESULTS The correlation between each subject s scores and the average of the rest of the group was calculated. The lowest correlation was just under 0.4 (Subject 10) and thus all subjects data were kept in the subsequent analysis. The standard deviation of the estimated mean of each sound-scale pair was no more than 7.5 % of the scale length for all but one scale ( Industrial-- Natural which was 9.6%). The standard deviation of the estimated mean for the majority of signal-scale scores was between 2.5 and 5% of the entire scale. Metric-Response Correlations Metrics were correlated to the responses, the corresponding R 2 values are shown in Table 2 for a subset of the scales and metrics. The highest correlations were found when using the maximum of the output of the time-varying loudness model developed by Moore and Glasberg (2002). Both forms of this model, the short-term and the long-term Loudness, were highly correlated. Much lower correlations were found when other metrics were used, even statistics of Zwicker s time-varying Loudness (Zwicker, 1998). Table 2. Coefficients of determination (R 2 ) for linear single metric models. Loud Soft Violent - Peaceful Acceptable - Not Acceptable' Annoying - Not Annoying Startling - Calming Stationary Loudness (ISO 532B) 0.684 0.522 0.542 0.405 0.408 Roughness [asper] (Zwicker & Fastl, 1998) 0.220 0.132 0.072 0.037 0.061 Zwicker Loudness Max 0.495 0.328 0.444 0.332 0.272 Zwicker Loudness exceeded 5% of the time 0.222 0.123 0.088 0.025 0.043 Zwicker Sharpness exceeded 1% of the time (S1) 0.281 0.186 0.058 0.037 0.081 Zwicker Sharpness exceeded 5% of the time (S5) 0.145 0.059 0.009 0.001 0.009 Moore Loudness Max 0.887 0.767 0.812 0.732 0.727 Moore Loudness exceeded 5% of the time 0.380 0.296 0.186 0.105 0.175 Moore Long Term Loudness Max 0.883 0.789 0.697 0.621 0.700 Moore s Long Term Loudness exceeded 5% of the time 0.854 0.758 0.654 0.570 0.655 A-weighted Sound Exposure Level (ASEL) 0.637 0.460 0.473 0.396 0.356 3
Loudness Rise Time 0.125 0.161 0.299 0.359 0.350 The high levels of low-frequency components in these signals may be one reason for the difference in performance between the Zwicker and Glasberg & Moore based Loudness metrics; it is in this region that the two models differ most. The mean of the responses on some of the scales plotted against selected metrics are shown in Figure 1. As expected, there is greater variability in the annoyance results, see, e.g., differences in responses to signals 6 and 21 which are the same signal, than the loudness and startle results, though for three out of the four repeated signals (22&19, 23&1, 24&13) the annoyance judgements are very consistent. Thunder, the filtered car door slam and the gunfire were rated as more annoying than other signals and this rating appears to be well explained by the maximum loudness of these sounds. Figure 1. Mean annoyance, loudness and startle scores plotted against various metrics. Estimated startle is a linear model of maximum Loudness and Loudness Rise Time. Signals 1-10,19f, low N-wave recordings; 11-12 and 16-18 simulated low-amplitude booms; 13f & 14, car door slams; 15, gunfire; 20, thunder; and 21-24 repeats of 6,19,1 and 13, respectively. f denotes a high-pass filtered signal (150 Hz cut-off frequency). Number of Sound Attributes To determine the perceptual dimensionality of the test stimuli, a factor analysis of the twelve word scale scores pertaining to perceptual dimensions (e.g. Loud/Soft, Sharp/Dull) data was performed. (The MATLAB program factoran was used with the promax option for factor rotation chosen.) The percent variance explained by each independent factor is shown in Figure 2. It appears that the number of strong factors is likely to be 3 or 4. The factor loadings for 3 common factors are shown in Figure 3. Three of the factor loadings when 4 common factors were selected were very similar to the loadings when 3 were chosen. The factors for the rotation shown in Figure 3, are roughly aligned with scales related to duration, startle and distance, the latter contributing strongly to both spectral balance and loudness scales. 4
Figure 2. Percent of variance explained by each common factor from an analysis of the eleven perceptual scales. 'No echoes--lots of echoes' DISCUSION 'Shallow--Deep' 'Clean--Rumbling' 'Short--Long' 'Calming--Startling' 'Distant--Close' 'Slow--Fast' 'Smooth--Harsh/Rough' 'Rounded--Peaked' 'Low pitch--high pitch' 'Dull--Sharp' 'Soft--Loud' -1-0.5 0 0.5 1 Figure 3. Factor loadings when choosing three common factors. From the linear fits between the individual word scales and metric, it appears that the maximum output of the time-varying loudness algorithm recently developed by Glasberg and Moore (2002) outperforms the two other metrics (statistics of Zwicker s time-varying Loudness and A- weighted Sound Exposure Level). This result differs somewhat from the previous research conducted by Sullivan (2004) in that the A-weighted Sound Exposure Level (ASEL) did not perform as well in this study. This may be due to the greater variety in stimuli in this experiment; gunfire, car-door slam recordings and thunder were also included as test stimuli. These additional sounds are dissimilar in temporal characteristics from sonic booms, namely they are not N-waves, and they also differ in spectral character. However, low-amplitude sonic booms will be judged in a context where these other types of transient sounds also occur. Therefore, it is important to find metrics that are appropriate for predicting the impact of a variety of transient environmental sounds, rather than ones that work well for only one type of transient. 5
From the factor analysis there appear to be 3 significant common factors. In the rotation of components shown here, these are aligned with duration, startle and closeness. Clearly loudness plays a role in startle and perceived distance. Spectral balance will also play a role in perceived distance because of the differences in high and low frequency attenuation with propagation distance (this is reflected by this factor s contributions to the spectral balance related scales see Figure 3). The rise time of the Loudness (time from background to maximum Loudness) also plays a role in startle and inclusion of it in a simple linear model with maximum Loudness yielded an 8% improvement over predicting startle response with maximum Loudness alone. CONCLUSIONS A semantic differential test was conducted to examine the perceptual space of transient sounds including sonic booms and to examine which metrics may be appropriate for evaluating this class of sounds. It was found that loudness and annoyance responses are more highly correlated to the maximum output of Glasberg and Moore s time-varying loudness model than to statistics of Zwicker s time-varying loudness model predictions or the A-weighted Sound Exposure Level, a metric that has often been used in evaluation of transient sounds. The better performance of the Glasberg and Moore s model may be due to the differences between it and Zwicker s model at low frequencies, because many of the sounds in the test have relatively high levels of low frequency energy. Two of the three dimensions of the sounds that were found in a factor analysis of the scores on the eleven perceptual scales are related to loudness. This test involved relatively few subjects (14) though mean ratings for duplicated signals were, in most cases, very close; future work will involve adding more subjects and improving models of common factors, annoyance and other scales related to overall judgements of the sounds. ACKNOWLEDGEMENTS The authors wish to thank the FAA/NASA/TC PARTNER Center of Excellence for their financial support. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the FAA, NASA or Transport Canada. The authors also wish to thank Vic Sparrow of Pennsylvania State University and Brenda Sullivan of NASA Langley, for their leadership as well as the Project 8 industrial and government partners who provided some of the recordings used in the tests. REFERENCES B. R. Glasberg and B. C. J. Moore. A model of loudness applicable to time-varying sounds. Journal of the Audio Engineering Society, 50(5):331 42, May 2002. J. D. Leatherwood, K. P. Shepherd, and B. M. Sullivan. A new simulator for assessing subjective effects of sonic booms. Technical report, NASA, Sept 1991. A. Marshall, P. Davies, B. Sullivan, and A. Pilon, Preliminary work on the development of a lexicon for supersonic aircraft sound, Journal of the Acoustical Society of America, 120: 3120-3121, 2006. A. Marshall, P. Davies, Perceptual attributes of low amplitude sonic booms and other transient environmental sounds heard outdoors, Proceedings of the 19 th International Congress on Acoustics, Madrid, Spain, Sept. 2007. C.E. Osgood and G.J. Suci. Factor analysis of meaning. Experimental Psychology, 50: 325 338, 1955. K. Plotkin and B. Bradley. The effect of onset rate on aircraft noise annoyance, Volume 1: Laboratory experiments. Technical report, Wyle Laboratories, November 1991. B.M. Sullivan. Human response to simulated low intensity sonic booms. In Proceeding of Noise- Con 2004, Baltimore, 2004. E. Zwicker and H. Fastl, Psychoacoustics: Facts and Models, 2 nd Edition, Springer-Verlag, 1998. 6