Boundary array size for teleconferencing

Abstract

Boundary arrays for teleconferencing require large numbers of sensors to be effective. The paper describes efforts to estimate the number of microphones required for an effective teleconference boundary array. Measurements of many source-to-sensor room transfer functions (RTF) were made in a conference room of moderate size. The direct arrivals of these RTFs were time aligned in software to estimate the transfer function of a delay-and-sum boundary array. Using the image method of Allen and Berkley, another set of RTFs was generated for a virtual room acoustically similar to the measured room. These RTFs were time aligned to create a second virtual array. The S/Rs of these two virtual arrays compare favorably. The image method was used to estimate transfer functions of virtual boundary arrays of different sizes for a 60-cubic-meter room. Standard beamforming and matched-filter processing were used. TIMIT speech was convolved with the virtual array transfer functions to generate audio used for evaluation. A listening test ranked the auditory performance of four single-microphone reference systems and eight test arrays in the virtual room. For rooms of roughly 100 cubic meters, boundary arrays with approximately 100 microphones may provide good subjective performance.