Empirical and modeled acoustic transfer functions in a simple room Effects of distance and

EFFECTS OF DISTANCE AND DIRECTIONBarbara G. Shinn-CunninghamBoston University Hearing

Research Center

Cognitive and Neural Systemsand Biomedical Engineering677 Beacon, Boston, MA 02215

shinn@cns.bu.edu

ABSTRACT

Empirical transfer functions were measured for a manikinhead as a function of source position (re: the listener) andlistener position (re: the room) for sources within a meter ofthe listener. Empirical results are compared to roomsimulations using a standard image-method model combinedwith anechoic, distance-dependent head-related transferfunctions (HRTFs). Results suggest that the biggestdiscrepancies between measured and modeled impulseresponses arise due to interactions of the head with thesource, which cannot be ignored for sources this close to thelistener. Results give insight into the importance of theacoustic effects of the head and room on the total signalreaching a listener and have implications for understandingspatial perception in rooms and developing realistic 3-Dspatial auditory displays.

1. INTRODUCTION

Echoes and reverberation (jointly termed “reverberation”throughout this paper) have many important perceptualeffects. Reverberation provides information about thelistening environment, causes slight degradations indirectional auditory acuity, vastly improves auditorydistance perception, and degrades speech intelligibilitycompared to listening in anechoic space. In virtual displays,inclusion of reverberation improves the subjective realism ofa display. While these results are well known, there is littlework quantifying how the physical effects of reverberationinfluence perception. The ultimate goal of this work is todevelop a room-acoustics model to enable quantitativeinvestigations of the perceptual influence of reverberation.Recent efforts in our laboratory have focused oncharacterizing the effects of reverberant energy on perceptionof source distance and direction [1-3], speech intelligibility[4], and other tasks, when sources are relatively near thelistener. For such source positions, robust interaural leveldifferences or ILDs change with source distance and lateralitydue to the interaction of the source with the listener’s head[5-7], and the distance-dependence of the direct soundcannot be accurately modeled by a simple 1/distance scaling.

In order to begin to identify physical bases for theseeffects in the signals reaching the listener, acoustic transferfunctions were measured for different source positions

Joseph G. DeslogeResearch Laboratory ofElectronics, MassachusettsInstitute of Technology

50 Vassar St

Cambridge, MA 02139jgd@alum.mit.edu

Norbert Kopão

Boston University Hearing

Research Center

Cognitive and Neural Systems677 Beacon, Boston, MA 02215

kopco@cns.bu.edu

relative to the listener and listener positions relative to theroom [8]. Empirical results are compared to results from aroom model that incorporates distance-dependent HRTFswith the simple image-method approach (e.g, see [9]). Thismodel is based upon a number of simplifying assumptions(e.g., the source is a uniformly-radiating point source, theonly reflecting surfaces are the six sides of the perfectly-rectangular room, the head interacts with each modeledreflection but not with the pattern of reflections, etc.).

2. METHODS

Acoustic transfer functions to the ear canal entrances weremeasured and modeled for sources at different positionsrelative to the listener’s head. Distances ranged from 15 cmto 1 m and azimuth angles from 0 deg to 90 deg (to the right).All sources were in the horizontal plane containing the ears.The measured / modeled room is roughly rectangular, withdimensions of 5 m x 9 m x 2.9 m; however, one long wall issemi-flexible and retractable, with fairly large absorption (allother walls were hard). Two listener positions wereinvestigated, with the listener positioned in either the centerof the room or the corner of the room (i.e., with back and leftside within two feet of the back and left walls) of the room.2.1. Empirical Measurements

Empirical measurements were made at the entrance to the earcanals of a Knowles Electronics manikin (KEMAR) using aMaximum-Length-Sequence (MLS; see [10]) of 32767 pointsat a sampling rate of 44.1 kHz. For the room in question, thislength was sufficient to measure the impulse responsebeyond the point at which the reverberant tail was 60 dBbelow the direct sound level.2.2. Room Model

Simulated impulse responses were generated using amodified version of the image method (e.g., see [9]). In thismethod, the reverberant room is modeled as an empty boxwith acoustically reflective walls, each of which has aspecific, frequency-dependent absorption characteristic.Reflecting the original source across the walls of the roomcreates image sources consistent with discrete echoes off thecorresponding wall. Each simulated source location is thenassociated with a measured anechoic HRTF-derived impulse

21-24 October 2001, New Paltz, New York

W2001-1

response. The measured impulse response is time shifted andamplitude scaled (preserving interaural cues) to account forsmall differences between the distance at which the impulseresponse was measured and distance of the image source.Finally, the impulse is filtered to account for the wallabsorption. All such image sources (computed iteratively outto the first 110 ms of the response, which is out to the pointat which empirical HRTFs were within 30 dB of their peakvalue) are summed to produce the total reverberant impulseresponse (no late reverberant tail was included). The currentsimulation differs from most previous, similar models in thatthe measured impulse responses for nearby sources are

distance- as well as direction-dependent.

3. RESULTS

3.1. Magnitude Spectra

Figures 1 and 2 show the measured and modeled(respectively) long-term magnitude spectra of the impulseresponse at the right (black) and left (gray) ears for differentsource positions when the listener is in the room center.

15 cm0corner of the room. In this case, frequency-to-frequencyvariations in the magnitude spectrum are also observed.However, strong, early reflections from the walls additionallylead to pronounced comb-filtering of the received spectra.This effect is strongest at the far ear; in some cases, the far-eardirect-sound impulse is lower in level than the first wallreflection. The resulting long-term magnitude spectra showstrong peaks and notches that arise directly from the in-phase and out-of-phase interference of early, discrete echoes.

15 cm0-20-40-60left earright ear100 cm45˚Magnitude Spectrum (dB)0˚0-20-40-60 100 cm00˚-20-40-60left earright ear90˚-20-40-60 1 510 151045˚Magnitude Spectrum (dB)Frequency (kHz)0Figure 2. Magnitude spectra of modeled transferfunctions for various source positions (room center).3.2. Interaural Time Differences

Interaural time differences (ITDs) were calculated from thelong-term spectra of the reverberant impulse responses.While such a simple analysis does not directly predict thedegree to which reverberation may degrade directionalhearing (since it ignores temporal structure and perceptualphenomena such as the precedence effect), it helps toquantify the degree of interaural distortion produced byreverberation. Figures 3 and 4 show ITD as a function offrequency for direct-sound-alone (gray) and reverberant(black) impulse responses for the center-room position.Figures 5 and 6 show corresponding plots for the corner-room position. ITD was calculated by dividing the interauralphase angle at each frequency by 2f; inherent phaseambiguity is shown by plotting the 2 multiples ofinteraural phase. The “true” ITD is roughly the ITD value ofthe resulting contour that is constant as a function offrequency.

In all cases, the interaural distortion is greatest for the farsources (right column) compared to the near sources (leftcolumn) and grows with source laterality. Empirical andmodeled results show similar patterns. For the center roomlistener positions (Figures 3 and 4), the ITD distortion isessentially random around the direct-sound-alone ITD. Forthe corner room listener positions (Figures 5 and 6), thedistortion is much more systematic, with large fluctuationsand discontinuities in the ITD as a function of frequency.This pattern arises from the comb-filtering effects of the early

-20-40-60090˚-20-40-60 1 510 1510Frequency (kHz)Figure 1. Magnitude spectra of measured transferfunctions for various source positions (room center).Qualitatively, measured and modeled results show thesame patterns. When the listener is in the center of the room,the main effect of reverberation is to cause frequency-to-frequency variations in the received spectral energy such thatthe magnitude spectra vary randomly about the direct-sound-alone spectral levels. The variations are largest in the left ear,which is, for the tested positions, always the ear farther fromthe source and the ear facing the “hard” wall in theasymmetrical room. These results are consistent with the factthat the dominant energy in the impulse response comesfrom the direct sound; the amount of spectral “distortion”depends on the direct-to-reverberant energy ratio (D/R),which decreases with distance, increases at the right ear anddecreases at the left ear as the source moves laterally to theright, and is larger in the right ear than the left.

While not shown here for sake of brevity, measured andmodeled results also agree well when the listener is in the

W2001-2

2001 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics

wall reflections and the frequency-dependent cancellationand reinforcement caused by the reflections.

115 cm100 cmdirect˚00.5reverberant01)s˚5m0.5(4 DT0I1˚00.590-0.50.511.520.511.52Frequency (kHz)Figure 3. Interaural time delay (ITD) versus frequencyfor measured transfer functions for various source

positions (room center).

115 cm100 cmdirect˚0.50reverberant01)s˚0.55m(4 DT0I1˚0.5090-0.50.511.520.511.52Frequency (kHz)Figure 4. ITD versus frequency for modeled transferfunctions for various source positions (room center).3.3. Direct- and Reverberant-Sound Energy

The qualitative agreement between the measured andmodeled impulse responses demonstrated in Figures 1-6 isencouraging; however, more quantitative comparisons arecritical for understanding how well the simple room-acoustics model describes the effects of reverberation. In allcases (for all physical cues considered), the amount of“distortion” produced by the reverberant energy variesdirectly with D/R. For source and listener positions yieldinglarge D/R, the reverberation causes small changes in thephysical signals at the ears; when D/R is small, thereverberation has a larger effect. Due to the fact that thedirect-sound HRTFs used in the simulation were taken from

21-24 October 2001, New Paltz, New York

empirically-measured, distance-dependent HRTFs, the directsound energy at the left and right ears changes identically inthe measured and modeled results. The distance-dependenceof these empirical HRTFs, which is usually modeled asvarying with 1/distance, differs from most other simulationsdue to the fact that sources are relatively close to the headand the interaction of the head and source wave cannot beignored.

115 cm100 cmdirect˚00.5reverberant0)1s˚5m(0. DT0I1˚00.590-0.50.511.5Frequency (kHz)20.511.52Figure 5. ITD versus frequency for measured transferfunctions for various source positions (room corner).

115 cm100 cm0.5direct˚0reverberant01)s˚5m0.5(4 DT0I1˚00.590-0.50.511.5Frequency (kHz)20.511.52Figure 6. ITD versus frequency for modeled transferfunctions for various source positions (room corner).One effect that is not perfectly captured by the model is asystematic increase in the reverberant energy in the measuredHRTFs (calculated by time-windowing out the direct impulsein the HRTFs and summing the remaining energy) withdecreasing source distance. Figure 7 shows the reverberantenergy reaching the left and right ears as a function of sourceazimuth (source distance varies parametrically in the figure).The amount of reverberant energy is almost always greater inthe left (dashed lines) compared to the right (solid lines) earas a result of the absorption characteristics of the right-side

W2001-3

wall. In addition to this, however, the reverberant energytends to decrease with source distance, particularly in theright (near) ear. We believe this pattern to be a result of theinteraction of the head and source. Specifically, for sourcesclose to the head, the head diameter is large compared to theradiating wavefront of the source. As a result, large amountsof energy are reflected back away from the head, adding newreflections not accounted for in the current room model.While our simulation includes the interactions of the headwith the source wavefront for the direct sound (including, forexample, the large interaural level differences that arise fornearby, lateral sources), it does not take into account thisinteraction in the reflectance model.

The model’s failure to accurately account for thedependence of the reverberant energy level on sourcedistance directly affects the modeled D/R. Since the D/Rdirectly predicts the magnitude of any acoustic effect of thereverberation on the total signals reaching the ears, this erroryields quantitative errors in the effects of reverberation onall perceptually-relevant spatial auditory cues.

-5-5.5Reverberant Energy(dB re: direct energy 0 deg, 1 m)leeft ararncefor nearby sources, failing to account for the effects of thehead on the reflected energy causes the most obviousdiscrepancy between measured and modeled results.

Future efforts directed towards including these effectsmay yield a room model that can be used to simulate realisticreverberation patterns for spatial auditory displays. Such amodel will be useful for exploring the importance ofreverberation for spatial auditory perception, particularly indistance, as well as for the effects of reverberation onperceived realism in spatial auditory displays.

5. ACKNOWLEDGEMENTS

This work was supported by the Air Force Office of ScientificResearch. Empirical data were measured by Tara Brown. WillCunningham helped in the preparation of this manuscript.For related materials, see http://www.cns.bu.edu/~shinn.

6. REFERENCES

[1]S. Santarelli, \"Auditory Localization of Nearby Sources

in Anechoic and Reverberant Environments,\"unpublished Ph.D. dissertation, Dept. of Cognitive andNeural Systems, Boston University, Boston, MA, 2000.[2]B. G. Shinn-Cunningham, \"Learning reverberation:

Implications for spatial auditory displays,\" Proceedingsof the International Conference on Auditory Displays,pp. 126-134, Atlanta, GA, 2000.

[3]B. G. Shinn-Cunningham, \"Distance cues for virtual

auditory space,\" Proceedings of the IEEE-PCM 2000, pp.227-230, Sydney, Australia, 2000.

[4]B. G. Shinn-Cunningham, J. Schickler, N. Kopão, and R.

Y. Litovsky, \"Spatial unmasking of nearby speechsources in a simulated anechoic environment,\" Journalof the Acoustical Society of America, vol. in press, 2001.[5]B. G. Shinn-Cunningham, S. Santarelli, and N. Kopão,

\"Tori of confusion: Binaural localization cues forsources within reach of a listener,\" Journal of theAcoustical Society of America, vol. 107, pp. 1627-1636,2000.

[6]D. S. Brungart and W. M. Rabinowitz, \"Auditory

localization of nearby sources I: Head-related transferfunctions,\" Journal of the Acoustical Society ofAmerica, vol. 106, pp. 1465-1479, 1999.

[7]R. O. Duda and W. L. Martens, \"Range-dependence of the

HRTF for a spherical head,\" IEEE ASSP Workshop onApplications of Digital Signal Processing to Audio andAcoustics, 1997.

[8]T. J. Brown, \"Characterization of Acoustic Head-Related

Transfer Functions for Nearby Sources,\" unpublishedM.Eng. thesis, Dept. of Electrical Engineering andComputer Science, Massachusetts Institute ofTechnology, Cambridge, MA, 2000.

[9]P. M. Peterson, \"Simulating the response of multiple

microphones to a single acoustic source in a reverberantroom,\" Journal of the Acoustical Society of America,vol. 80, pp. 1527-1529, 1986.

[10]D. D. Rife and J. Vanderkooy, \"Transfer-function

measurement with maximum-length sequences,\"Journal of the Audio Engineering Society, vol. 6, pp.419-444, 19.

rig eht-6d0.150.401aist-6.5-7-7.5-8-8.5-90204060Azimuth Angle (deg)80-9.5Figure 7. Reverberant energy in measured impulseresponses to the left and right ears as a function ofsource direction and various distances (room center).

4. CONCLUSIONS

Direct and reverberant energy levels vary systematically withsource distance as well as direction when sources are near alistener. As a result, D/R also varies systematically for nearbysources in a room. Since this energy ratio determines howlarge the acoustic influence of reverberation will be (i.e., howreverberation influences the magnitude spectra and interauraldifferences of the signals reaching the ears), it is important tomodel this dependence appropriately. A simple room-imagemodel that incorporates the distance-dependence of thedirect sound impulse response yields qualitatively similarresults to empirically-measured impulse responses in asimple room; however, by ignoring the effect of the head onthe reverberation pattern, the approach fails to quantitativelyreproduce the dependence of reverberation (and the direct-to-reverberant energy ratio) on source distance. Othersimplifications in the model may also be critical; however,

W2001-4

2001 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics