Page  114 ï~~A Binaural Recording and Playback System for the Reproduction of Virtual Concert Halls Gilbert A. Soulodre Dale R. Stammen Faculty of Music, McGill University Faculty of Music, McGill University Abstract Many of the acoustic variables which govern the subjective quality of concert halls are, for the most part, poorly understood. Acoustical engineers have numerous objective measures at their disposal, but many of these require further correlation with subjective data to maximize their usefulness. A technique which offers an excellent opportunity for correlating subjective and objective measures is the binaural recording-playback system. While this technique has been in use for some 30 years, current computing capabilities make its use more feasible. We present a computer-based system that measures the binaural acoustic impulse response of a concert hall and convolves the impulse response with anechoically recorded music. The convolved music is then played through a binaurally accurate reproduction system creating a virtual representation of the original concert hall. This form of virtual hall allows different halls, as well as different positions in a hall, to be compared on a pairwise basis. To date binaural impulse responses from more than a dozen concert halls across North America have been captured and convolved using our system. The system also allows impulse responses to be analyzed and edited in software. This permits certain acoustic parameters to be held constant while the subjective effects of others are investigated. Our system allows composers of electroacoustic and computer music to convolve (and audition) their music with impulse responses of actual concert halls. The binaural recording-playback system with its library of concert hall impulse responses offers researchers and composers a valuable tool for their work. 1. Introduction In this paper we describe a system for the measuring and processing of binaural impulse responses of concert halls. The system further allows the user to convolve music with these impulse responses and audition the resultant sound fields in a binaurally accurate manner, thus creating a form of virtual concert hall. Development of the system was motivated by the need to conduct A/B listening tests of various concert halls. The system consists of three main parts: data acquisition, signal processing, and playback. 2. Measurement of binaural impulse responses Binaural impulse responses are obtained using the RAMSoft system developed at the National Research Council of Canada [Halliwell and Bradley, 1989]. RAMSoft is an IBMPC based acoustic measurement system which uses maximum length sequences (MLS) to compute the impulse response of a concert hall. From these impulse responses various acoustic measures such as reverberation time (RT-60), early decay time, lateral energy fraction (LF), and interaural cross-correlation (IACC) can be calculated. To capture the binaural impulse responses a sound source is placed on the stage of the concert hall and a receiver (a dummy head with microphones at the eardrums) is placed at an appropriate listening position in the audience seating area. The impulse response from source to receiver is then measured. The sound source consists of a dodecahedron (12 sided) loudspeaker, which provides an omni-directional radiation pattern across a wide range of frequencies. The dummy head used in this system is a Briiel & Kjaer Type 4128 Head and Torso Simulator. Several source-receiver positions are measured within each hall to ensure that a proper representation of the hall's acoustics is obtained. All of the impulse responses are 65535 samples in length and were sampled at a rate of 32 kHz. To date a database of over one hundred measurements in more than a dozen major North American concert halls has been collected. Table 1 lists the halls measured by our system. 7A.3 114 ICMC Proceedings 1993

Page  115 ï~~Boston Symphony Hall, Boston, Massachusetts Massey Hall, Toronto, Ontario Mechanics Hall, Worcester, Massachusetts Orchestra Hall, Detroit, Michigan Saunders Theater, Cambridge, Massachusetts Academy of Music, Philadelphia, Pennsylvania E.J. Thomas Performing Arts Hall, Akron, Ohio Kleinhans Music Hall, Buffalo, New York Meyerhoff Symphony Hall, Baltimore, Maryland Troy Music Hall, New York, New York Centennial Concert Hall, Winnipeg, Manitoba Severance Hall, Cleveland, Ohio Western Manitoban Centennial Auditorium, Brandon, Man. J.F. Kennedy Center, Washington D.C. Table 1: Measured Concert Halls 3. Processing of the Binaural Impulse Responses We have developed a software package called Sound Utilities which provides all of the analysis, editing, processing, and playback capabilities of our system. The software is is currently running on a Macintosh Quadra 800 computer. We are using a Digidesign Audiomedia II card for audio processing and playback. Sound Utilities accepts impulse responses in RAMSoft, Maximum Length Sequence System Analyzer (MLSSA), or Sound Designer II (SDII) formats and outputs binaural sound files in SDII format. We use the following procedure to create a binaural sound file from a concert hall impulse response. 1. The left and right channels of the hall impulse response are converted from RAMSoft to Macintosh format. Sound Utilities then transforms the impulse responses into the frequency domain and saves them as HALL files. Several impulse responses may be converted at once by creating a conversion list. Amplitude scaling may also be added to selected impulse responses in order to adjust for amplitude differences resulting from the acquisition process. A sample of a concert hall impulse response is shown in Figure 1. Figure 1: Impulse Response, Boston Symphony Hall Seat H30 - Left Ear 2. The selected source material (anechoic music, electronic music, or noise) is preprocessed by convolving the source with the inverse of the diffuse field response of the dummy head. This head-corrected source material is then saved as a MUSIC file. This conversion process is shown in Figure 2. At this stage, the impulse responses of the crosstalk cancellation and speaker frequency response correction required by the binaural playback system are converted into PATH format files. The output of the MUSIC/HALL convolution will be post-processed by these PATH files. Our program currently uses four PATH files to specify the speaker correction and crosstalk cancellation. The measurement of these impulse responses will be discussed below in Section 4. Bruckner.ane choic H AT S.T IM.HE AD ~ B ruckn Figure 2: Conversion of Source Material to MUSIC ICMC Proceedings 1993 115 7A.3

Page  116 ï~~3. The user creates a convolution list specifying the source MUSIC file, the HALL file, the PATH correction files and a SDII output file. Sound Utilities then performs the specified convolutions. The MUSIC file is first convolved with the HALL impulse response. The left channel of the MUSIC/HALL convolution is then convolved with the left speaker frequency correction (PATHL1) and the left channel crosstalk cancellation (PATHL2). Similarly, the right-channel output of the HALL convolution is convolved with PATHR1 and PATHR2. The convolved paths are then recombined, normalized and converted into a binaural SDII file. The signal-processing paths are shown in Figure 3. On a Macintosh Quadra 800, using a two-second long HALL impulse response, each minute of stereo output takes approximately 15 minutes to compute. This performance is expected to improve as we add more RAM to our system and as we optimize our convolution algorithms. left speaker correction O-N P ATHL 1 left crosstalk cancellation MUSICHALL ___ P ATHL2 BOSTNDOO.IR.HALL PAT Bruckner.conv.SDA PATHRI right speaker correction PATHR2 right crosstalk cancellation Figure 3: Signal Processing Paths The current version of Sound Utilities also provides for the display and editing of impulse responses. and binaural sound files. For example, impulse responses may be trimmed in length, thus allowing for the relative contribution of the early reflections and the reverberant tails of an impulse response to be studied in a systematic way. Mixing of sound files, including automated amplitude and panning control, is also supported. We have found this feature to be particularly useful in convolving the two channels of a stereo source with two RAMSoft binaural impulse responses measured with the source at stage left and stage right. The resultant sound files are then mixed into a single binaural sound file. This results in a more realistic reproduction of the source-material over the binaural playback system. 4. Reproduction of Binaurally Convolved Music The final task in the process is to reproduce the sound files while remaining faithful to the binaural characteristics of the original concert hall. Loudspeaker playback was chosen over headphone reproduction because it tends to be inherently more effective since the listener's entire auditory mechanism is used as part of the process. In a binaural reproduction system it is fundamental that the left ear receives only the signal from the left channel and the right ear receives only the signal from the right channel. However, in normal stereo loudspeaker reproduction a form of crosstalk occurs wherein each ear receives part of the signal that was intended for the other ear. This interaural crosstalk is eliminated using mechanical 7A.3 116 ICMC Proceedings 1993

Page  117 ï~~barriers to minimize acoustic leakage from each loudspeaker to the opposite (contra-lateral) ear and by processing the sound files to cancel the remaining (mostly low frequency) crosstalk. Using MLSSA, we captured the impulse response of the crosstalk cancellation required for the left and right speakers. We also captured the free-field response of the loudspeakers corrected to obtain a flat frequency response. These impulse responses were then converted into PATH files as described above, and were used to post-process the output from the MUSIC/HALL convolutions. The binaural sound files created by the Sound Utilities program may be played individually or on a pairwise basis. Several binaural sound files may be combined into a play list that allows for convenient auditioning of each sound file. Play lists may also be used to group files together for automated double-blind paired comparison testing. This allows for subjective testing that exhaustively compares specific features of a variety of concert halls. The program allows the listener to switch back and forth between two sound files and records their selection to a data file. We have used this feature of the program to investigate relationships between LF and IACC [Soulodre et al., 1993A] and to subjectively compare Boston Symphony Hall and Massey Hall [Soulodre et al., 1993B]. 5. Other Applications The Sound Utilities program was designed as a general-purpose tool for manipulating sound files. The program offers composers the opportunity to audition their music in a variety of seats or halls using our library of concert hall impulse responses. For example, Sound Utilities was used by one of the authors to convolve sound files, add reverberation, and mix sound files for the composition tuva! [Stammen and Terriah, 1993]. 6. Future Work One limitation of the present playback system is that it does not adapt for head movements by the listener. It is believed that the addition of head-tracking capabilities will serve to improve front/back localization [Kendall, 1992]. The validity of our binaural system as a research tool has been confirmed through a series of parallel experiments using multiple loudspeakers in an anechoic chamber to create simulated sound fields [Bradley et al., 1993]. Future developments to Sound Utilities will include automated calculation of acoustic measures such as RT-60, LF, IACC etc. from the impulse response, as well as the creation and editing of unique impulse responses from entered parameters. References [Bradley et al., 1993] John S. Bradley, Gilbert A. Soulodre, and Neil Popplewell. Pilot Study of Simulated Spaciousness. Presented at the 125th meeting of the Acoustical Society of America, Ottawa, 1993. [Halliwell and Bradley, 1993] R.E. Halliwell and J.S. Bradley. RAMSoft II: A Computer-based Room Acoustics Measurement System. Journal of the Acoustical Society of America, 89(4), p. 1897. [Kendall, 1992] Gray S. Kendall. Directional Sound Processing in Stereo Reproduction. Proceedings of the 1992 International Computer Music Conference, pp.261-264, 1992. [Soulodre et al., 1993A] Gilbert A. Soulodre, John S. Bradley, and Dale Stammen. Spaciousness Judgments of Binaurally Reproduced Sound Fields. Presented at the 125th meeting of the Acoustical Society of America, Ottawa, 1993. [Soulodre et al., 1993B] Gilbert A. Soulodre and John S. Bradley. A Subjective Comparisons of Massey Hall and Boston Symphony Hall. Presented at the annual meeting of the Canadian Acoustical Association., Toronto. (Stammen and Terriah, 1993] Dale Stammen and Sean Temrah. tuva! for saxophones, MID! electric guitar and electronic instruments. Unpublished music manuscript, 1993. ICMC Proceedings 1993 117 7A.3