Proceedings of Meetings on Acoustics

Proceedings of Meetings on Acoustics Volume 19, 2013 http://acousticalsociety.org/ ICA 2013 Montreal Montreal, Canada 2-7 June 2013 Psychological and Physiological Acoustics Session 4aPPb: Binaural Hearing (Poster Session) 4aPPb25. Validating a binaural head for use in jury testing Jeremy Charbonneau*, Colin Novak and Helen Ule *Corresponding author's address: Mechanical Engineering, University of Windsor, 401 Sunset Avenue, Windsor, N9B3P4, Ontario, Canada, charbo6@uwindsor.ca A test procedure for use in loudness perception tests must be created to completely describes a phenomenon while at the same time minimizing jury listening fatigue. One contributor to this fatigue is the amount of time necessary for the test subject to experience all the required signals. Head and torso simulators have been used for years as a means to reliably quantify the acoustic performance of a product while avoiding the influence of listener bias and fatigue. This procedure not only controls the test parameters but also removes any human error that may occur. The purpose of this investigation is to qualify a head and torso simulator for use in loudness investigations. The objective of this experiment is to correlate the results from using this equipment to human subject results for high resolution experiments on directionality of loudness. Published by the Acoustical Society of America through the American Institute of Physics 2013 Acoustical Society of America [DOI: 10.1121/1.4799981] Received 27 Jan 2013; published 2 Jun 2013 Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 1

INTRODUCTION A key element in the documentation of an experiment is the collection of a sufficient sample of data for statistical analysis and determination of repeatability. However, when human subjects are involved, the amount of data available is often limited by the time it takes for participants to complete the test as well as psychological factors, such as participant fatigue and state of mind. For the field of psychoacoustics, human participants are often used as jurors to investigate the various perception characteristics of sounds. These measurements often require specific feedback from a participant where the amount of time required for a high resolution of data can be extensive. Investigations that require prolonged listening can be negatively skewed as participant fatigue increases and perceptions vary. When exposed to a particular sound repeatedly for example, it is possible for the subject to develop an acoustic memory, which may influence subsequent decisions. To confront this, head and torso simulators are often used instead of human subjects as an attempt to remove the listener bias as a source of error. The author s research is primarily focused on the perception of loudness from stationary signals, such as random sounds and pure tones. Past participants have been asked to match the loudness level of reference and target signals in an attempt to further understand the characteristics of loudness. For a participant, completing the stationary loudness matches can be challenging however, non-stationary or time-varying samples are significantly more complex. Variations in sound pressure level, spectral content and temporal characteristics can cause the participants to have trouble relating an entire segment of sound to a single reference value. It is well documented that sound pressure levels alone do not allow you to easily quantify a psychoacoustic concept. The actual perception often includes a combination of factors which influence the listener. It is difficult to isolate loudness from all the other percepts including annoyance, roughness, sharpness and tonality, which vary from person to person. By replacing the human participant with a binaural head, these psychological factors are removed, leaving the raw data for analysis. From this, it is then possible to collect consistent and repeatable results during an experiment where listening can be calibrated to ensure accurate measurements each time. The end goal of this research is to develop a future experimental method to investigate time-varying loudness. At present there are two time-varying loudness metrics which are being considered for standardization: (i) the Deutsches Institut für Normung (DIN) standard DIN 45631/A1:2010-03, an amendment to the stationary Zwicker method; and (ii) the Glasberg and Moore Time-Varying Loudness (TVL) model, an extension of the stationary loudness model (Glasberg and Moore, 2002). Each of these models is capable of using a dataset from a single channel and calculating either a monaural loudness prediction or a binaural prediction based on a dichotic assumption. EXPERIMENTAL SET-UP An experiment was created to validate the use of a binaural head to support jury test results and gain a higher resolution of detail. In order to obtain this, the loudness matching capability of a participant was assessed at varying listening angles which were further refined using a high resolution approach from a binaural head. A listening test was conducted in a semi-anechoic chamber where participants were asked to perform a loudness matching task. A Totem Acoustic MANi-2 loudspeaker, chosen for its wide frequency range capability, was located 2 metres (m) in front of the seated participant where five (5) target sounds were paired with a reference 60 decibel (db) white noise sample. The five test signals included: (i) pink noise; (ii) a complex multi-tone signal with sinusoidal waves including 125Hz, 250Hz, 500Hz, 1kHz, 2kHz, 4kHz, and 8kHz; (iii) a 1 khz square wave; (iv) a constant speed circular saw recording; and (v) a constant speed electric motor recording. Each signal was 4 seconds in length and included a cosine ramp-up and ramp-down function to remove any audible clicks on start-up. Due to the difficulty of matching non-stationary signals, each signal selected for this study was either stationary or quasi-steady. The primary goal of this investigation was to determine how an equal loudness response would vary with respect to incidence angle. The results were subsequently compared against those of a binaural head with the same test signals presented at a constant sound pressure level. Four of the samples used in this investigation were manufactured sounds, which a person would not normally be exposed to in daily interaction. A recent study by Ellermeier et al suggested that the meaning of a noise source may positively or negatively impact a subjects perception of it, while having little to no effect on the perceived loudness level (Ellermeier et al, 2004). Their results indicated that by removing the meaning of the signal, specific attributes such as annoyance and tonality were altered in the participants perception, overall changing their opinion Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 2

of the signal. As the loudness of the signal was not influenced, it is anticipated that the manufactured sounds and the recorded signals in this study would produce similar results. Jury Test Procedure Nine (9) human subjects took part in the study; six males and three females, all between the ages of 20 and 32 years of age. Each signal was presented to jury subjects as a two channel audio file. The first signal contained the reference white noise at a constant sound pressure level of 60 db and the second contained the target test file. Before being presented to the participants, the two channels were combined electronically through a mixing console and delivered through the common loudspeaker. The audio files were presented as an eight second loop, alternating continuously between the reference and the target channels. The loop continued until the participant indicated satisfaction that the two signals were of equal loudness. The participants were placed in a seated position, 2 m in front of the loudspeaker as shown in Figure 1 below. The chair was placed on a large rotating platform capable of rotating 360 degrees at any increment. To limit the amount of time that participants were required to be in the room only select angles were targeted including: -90 (with the participant facing left), -45, -15, 0 (looking straight ahead), 15, 45, and 90 (facing right). Subjects were asked to listen to the looping audio file and adjust the amplitude of the target signal until the two samples were perceived to be equally loud. In order to do this, the amplitude of the target signal was completely controlled by the participant via a 10-turn potentiometer. Once matched, the participant gave indication that they were satisfied at which time a 16 second recording was made of the presentation using a microphone located above the participants head. These recording files were used as a method of documenting the participant matches and subsequent analysis. FIGURE 1. PlanView of the Experimental set-up in the semi-anechoic room. To control the participants head position as the chair was rotated; visual cues were placed on the walls of the semi-anechoic chamber at specific angular locations. Participants were instructed to sit with their feet flat on the floor and with their head looking straight ahead towards the appropriate marker. Head and Torso Simulator Trial Once the jury tests were completed, the results were averaged to determine the sound pressure level of perceived equal loudness. In an attempt to understand how each signal was influenced by incidence angle, each signal was presented to a Brüel and Kjær 4100D head and torso simulator (HATS) at a constant amplitude as the assembly was turned. The HATS device was positioned in the same chair for consistency with the head forward as was instructed during the jury test. Likewise, the ear height was raised to 1.2 m (in the seated position), consistent with the average ear height measured from the human subjects that took part in the study. The entire set-up was rotated the full 360 using the rotating platform as recordings were made at 10 increments. Information was collected via the left and right channels of the HATS while the constant amplitude of the signal was monitored using the external microphone located above the chair. Much like with a person, the torso of the HATS will absorb and reflect different frequencies of sound, resulting in the inherent head related torso function (HRTF) into the signal. As each loudness metric approximates its own version of the head related torso function, the 4100 HRTF was filtered out of each ear s recording using a sound Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 3

quality software suite prior to analysis. Once removed, mono-wav files were created of each channel and the subsequent time-varying loudness was calculated using the DIN and TVL loudness metrics. RESULTS Prior to each experiment, the participants were given a hearing test to determine their threshold levels of perception for both their left and right ears from 20 Hz up to 20 khz. Two of the participants were found to have moderate hearing loss and their data was monitored closely to ensure that it provided results consistent with the others. Jury Test Results Participants reported no difficulty in completing the loudness matching procedure with the exception of matching the octave band and the square wave signals; each of which were characterized as annoying by nearly all of the participants. This is also likely attributed to the resulting variance of the subsequent responses. The reference white noise when presented at 60 decibels resulted in a loudness level of 78 phons. Should the loudness models function the way they are intended to, each matched pair should produce a flat response of equal loudness as the source and target signals were always presented from the same direction. Figures 2 and 3 below illustrate the analyzed loudness matches for both the DIN 45631/A1 loudness model and the TVL Model. FIGURE 2. Jury test results analyzed using the DIN 45631/A1 loudness model. Clockwise from the top left, plots of equal loudness between the reference white noise and: (a) the pink noise, (b) octave tones, (c) 1kHz square wave, (d) electric motor noise, and (e) circular saw noise. Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 4

FIGURE 3. Jury test results according to the Time-Varying Loudness (TVL) model. Clockwise from the top left, plots of equal loudness between the reference white noise and: (a) the pink noise, (b) octave tones, (c) 1kHz square wave, (d) electric motor noise, and (e) circular saw noise. The outcomes of the jury tests were similar to previous studies done by the authors. The participants were able to consistently match the loudness of the target signal to the reference noise, regardless of the listening angle (the reference signal and the target signal were always presented from the same direction). This result is clearly illustrated by the flat horizontal lines on the incidence plots. The shaded areas represent the range of values from the various participants both above and below the average. The narrow spread indicates a consistent perception among the participants. The large spread of responses for the octave tones, square wave and circular saw noise shows a clear indication of the difficulty that the subjects had in recognizing a match between the target and reference signals. Participants who indicated signs of slight to moderate hearing loss produced results consistent with the rest of the subjects; therefore a decision was made to include their results in the pool of collected data. The results of the averaged jury test are summarized in Table 1 below. From the sound pressure levels, an accurate match was made for the wide spectrum signals (Pink, Saw, and Motor).The participants were in agreement that these sounds were the easiest to evaluate and match. The average loudness levels however, indicate that the results for the circular saw and the electric motor were slightly below expected, regardless of the matched sound pressure levels. The octave bands and square wave were both more difficult to match and had significantly different loudness levels than expected. TABLE 1. Perceived equal loudness response, averaged from all jury test participants. Comparisons were made against a 60 db White noise signal presented from the same incident listening angle as the target. Target Signal Measured Sound DIN 45631/A1 TVL Loudness Pressure Level for Loudness Level Level Matched Pair White Noise (Reference) 60 db 78 phons 79 phons Pink Noise 58 db 79 phons 79 phons Octave Band Tones 52 db 68 phons 70 phons 1 khz Square Wave 51 db 61 phons 65 phons Circular Saw Recording 58 db 76 phons 79 phons Electric Motor Recording 59 db 76 phons 77 phons Following the jury experiments, the averaged perceived sound pressure levels for each loudness match was collected for the binaural head presentation. All six signals including the reference white noise were presented to the Type 4100D HATS at the fixed amplitudes, for an entire rotation of 360. Head and Torso Simulator The use of the HATS mannequin allowed for a more in-depth investigation of loudness measurement at various angles of incidence compared to the use of jurors who can become fatigued relatively easily. Recordings from both the left and right channels were made at 10 increments, allowing for a detailed plot of each-time varying loudness Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 5

metric as shown in Figure 4 and Figure 5. Each plot clearly indicates how the measured loudness level varied as the binaural head was rotated. In order to remove the need for separate scales, each plot was normalized to its own respective average level and is presented on a common deviation scale from -10 phons to +10 phons. The response of the DIN loudness model in figure 4 resulted in a smooth polar plot where the measured levels increase and decrease as each ear moves in and out of the heads shadow zone. Each of the signals showed a small fluctuation in loudness when either ear was facing the opposite direction as the incident source (+/-90 ). Overall, the response of the DIN standard to the binaural recording correlated well to the smooth characteristics of the observed sound pressure levels. Minor fluctuations were observed for the octave bands and the square wave while transitions remained smooth. The plots in Figure 5 illustrate the high-resolution response capability of the TVL metric to the constant sound pressure levels. Each plot shows sharp transitions between measurement points and notably higher loudness level to the right and left of full frontal incidence (0 ), as the head is turned to either side. The difference between Figures 4 and 5 are immediately apparent. While the DIN 45631 Amendment produced a smooth response as the head was rotated, the TVL response was much more drastic with sharp changes in predicted loudness level. The cause for this may be due to the post processing required for the separate metrics. The DIN timevarying loudness model had the advantage of being included with the acquisition software package, and therefore, did not require additional signal processing. The signals were taken from the acquisition software which also applied the inverse HRTF filter. This was subsequently inserted into the sound quality post-processing software which calculated the time-varying loudness for the left and right ears simultaneously. The TVL model required additional signal processing which may have negatively influenced the data. Once the HRTF was removed, single channel WAV files were exported and subsequently resampled using the accompanying executable file. Once resampled, each data set had to be recalibrated using a recorded calibration tone taken at the time the experiment was run. As a limitation to the procedure, only one calibration point could be used to calibrate the entire rotation of the binaural head. As a necessary assumption, it was assumed that all recordings made during each data set would require the same gain adjustments. FIGURE 4. Binaural head results analyzed using the DIN 45631/A1 time-varying loudness model. Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 6

FIGURE 5. Binaural head results analyzed using the TVL time-varying loudness model. While the TVL software required these additional steps, the symmetry of the white noise, pink noise, square wave and electric motor noise suggest that the TVL model produced a correct response to the data and that the equipmental set-up was consistent. While the unsymmetrical response of the circular saw and octave bands suggests an error in the acquisition, the same information was analyzed for both Figure 4 and Figure 5. The exporting and resampling of the data for the TVL model may have altered the data in an unintended way prior to analysis; should a more streamlined analysis be available, it may remove these inconsistencies. As part of the investigation into timevarying loudness, improvements to each model will be looked into with future research. In order to compare these results with the jury test findings, the specific incident angles used above are highlighted in Tables 2 and 3 below. TABLE 2. Resulting DIN 45631/A1 monaural time-varying loudness levels (phons) from a binaural head at select incidence angles for a comparison to a jury subject response. Target Signal -90-45 -15 0 15 45 90 Listening Channel L R L R L R L R L R L R L R White Noise (Reference) 82 93 85 94 89 92 91 91 92 89 94 84 93 83 Pink Noise 83 92 84 93 88 91 90 90 91 88 93 84 92 83 Octave Band Tones 72 80 75 82 80 81 82 81 82 78 82 71 80 71 1 khz Square Wave 69 77 70 78 74 78 76 76 78 74 78 68 76 67 Circular Saw Recording 81 91 83 92 87 90 89 89 90 87 92 82 91 81 Electric Motor Recording 77 86 78 86 81 84 83 83 84 81 86 77 86 77 Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 7

TABLE 3. Resulting TVL Model monaural time-varying loudness levels (phons) from a binaural head at select incidence angles for a comparison to a jury subject response. Target Signal -90-45 -15 0 15 45 90 Listening Channel L R L R L R L R L R L R L R White Noise (Reference) 80 81 81 81 77 80 79 79 80 77 81 81 81 80 Pink Noise 80 81 81 80 85 80 78 78 80 85 80 81 80 80 Octave Band Tones 79 86 81 78 85 86 78 85 78 84 78 78 86 79 1 khz Square Wave 75 83 77 84 80 84 82 82 83 80 84 75 82 75 Circular Saw Recording 78 78 79 79 83 77 76 82 83 80 84 75 82 75 Electric Motor Recording 83 83 83 82 78 81 80 80 81 78 82 83 83 83 While the loudness of the reference noise may vary as the head is rotated, the jury results have verified that the matched pairs have an equal loudness level to that of the reference signal, regardless of the incidence angle. From the tabulated values, it quickly became apparent that, although the TVL responses are more sporadic when plotted, the levels are consistently closer (never exceeding 8 phons from the anticipated value), to the white noise reference loudness. The DIN contours were smoother, consistently following the shape of the reference white noise level; thus producing the flat response seen in the jury test results. This supports a conclusion that the head and torso simulators may be used in place of human test subjects when the target loudness level is known; a value which may be determined through only limited jury test results. As was illustrated in the polar plots, the predicted loudness levels of the DIN model are seen to vary as much as 16 phons from the predictions based on matched loudness pairs. Therefore, while the DIN model produced smoother transitions with respect to incident angle, the accuracy of the TVL model may suggest that the TVL s response is more appropriate. If the resampling and calibration procedure of the TVL model could be improved, the binaural trends of the TVL model could be improved to potentially correct the issue with specific listening angles. CONCLUSIONS Five target signals were compared via a matched pair loudness experiment to a 60 decibel white noise sample. Test results were compared against the measured values from a binaural head. The end goal of this research was qualify the use of head and torso simulators for the investigation of loudness. It was determined that the binaural head response produced similar trends when using the streamlined DIN 45631/A1 approach. The results accurately followed the trends of the jury test results where only small fluctuations of equal loudness were seen as the listening angle was changed. It is concluded that head and torso simulating (HATS) devices can produce accurate results in the place of jury test participants as long as the presentation levels are known. For extensive psychoacoustic testing, these values may be obtained through limited jury test studies of small sample size. The Glasberg and Moore s TVL model demonstrated sharp fluctuations in loudness as the direction of the HATS device was rotated. It was determined that the most likely cause for these deviations was the post processing procedure requiring additional steps such as resampling and calibration of each loudness signal. Should this algorithm be included in a software package where WAV file information is taken automatically, it is believed that the TVL response would be more smooth, similar to the DIN polar plots. It should be reinforced that the matching procedure presented was not an investigation of loudness directionality. Such a study would require that the reference signal remain constant at a single incidence angle while the incidence of the target signal be varied, (for example see Sivonen, 2005). This investigation allowed the participants to match a reference signal to a target tone from various listening angles to determine the influence of source direction. Results indicated an ideal response to both the reference signal and the target signal as the results were equally impacted at each listening position. Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 8

FUTURE WORK It is planned that the author will revisit the data using individual HRTF s derived for all angles of listening and apply the resulting filters to the dataset. Using a binaural head, a high detail HRTF of the 4100D and other binaural heads will be explored in an attempt to further improve on this measurement and calculation procedure. Specifically, it will be determined how the loudness models are influenced by the HRTF if it is removed or left intact for a particular signal. Potential improvements for each time-varying loudness model may include a listening angle option when using a binaural head, such as for applications in product development or workplace exposure. ACKNOWLEDGEMENTS The author would like to thank Brüel and Kjær for their in-kind support of the equipment used in this experiment and their assistance in use of the analysis software. BIBLIOGRAPHY B.R. Glasberg and B.C.J. Moore. (2002). "A model of loudness applicable to time-varying sounds," Journal of the Audio Engineering Society, 50(5), 331-42. Brüel & Kjær Head and Torso Simulator Type 4100D, (Last accessed December 2012), available from: http://www.princeton.edu/3d3a/manuals/larsdatasheet.pdf. Deutsches Institut für Normung, (German Institute for Standardization). (2010). DIN 45631/A1:2010-03- Calculation of loudness level and loudness from the sound spectrum - Zwicker method - Amendment 1: Calculatoin of the loudness of timevariant sound, (German Institute for Standardization). Ellermeier, W, A Zeitler, and H Fastl. (2004). "Predicting Annoyance Judgements From Psychoacoustic Metrics: Identifiable Versus Neutralized Sounds." (The 33rd International Congress and Exposition on Noise Control Engineering). Sivonen, Ville Pekka. (2005). Directional loudness perception the effect of sound incidence angle on loudness and the underlying binaural summation, (Aalborg Univerisity). Proceedings of Meetings on Acoustics, Vol. 19, 050163 (2013) Page 9