A Study on Reference Microphone Selection for Multi-Microphone Speech Enhancement

Multi-microphone speech enhancement methods typically require a reference position with respect to which the target signal is estimated. Often, this arbitrarily chosen as one of microphones. However, it has been shown that choice microphone can have significant impact on final noise reduction performance. In paper, we therefore theoretically analyze selecting performance near-end being taken into account. Following generalized eigenvalue decomposition (GEVD) based optimal variable span filtering framework, find for any linear beamformer, output signal-to-noise ratio (SNR) taking both and far-end account dependent. Only when neglected, SNR rank-1 beamformers does not depend position. in general rank-r r > 1 (e.g., multichannel Wiener filter) Based these, propose an algorithm selection maximizes SNR. addition, lower-complexity still beamformers, but sub-optimal rank beamformers. Experiments using simulated array validate effectiveness proposed show terms quality, several dB be gained by proper microphone.