Microphone cross-array beamformer processing to reduce noise and reverberation

Abstract

Hands-free audio communication systems that are designed to allow audio and speech communication between remote parties are known to be sensitive to room reverberation and noise when the source is distant from the microphone. A common solution to the problem of room reverberation is to use an array of microphones to spatially filter the acoustic field. The maximum array gain is only attainable with specific microphone geometries and the gain of realizable microphone arrays is typically significantly lower than the maximum. The algorithm described in this talk attempts to address the rather slow growth in linear processing directional gain as a function of the number of microphones. One possible approach to attain higher directive gain is to use a processing scheme based on nonlinear multiplicative processing. An obvious candidate is the coherence function since it is a bounded and normalized multiplicative measure. A technique is presented for reverberation reduction processing using at least two beamforming microphones and a function of the estimated short-time coherence between the beamformer outputs. Each beamformer should have a different directional response or spatial position, or both, but with overlapping responses in the direction of the desired source.