Wiener filter design by estimating sensitivities between distributed asynchronous microphones and sound sources

A Wiener filter design method for distributed asynchronous microphones is proposed. The major approach to speech enhancement with a distributed microphone array has been to form directivity after adjusting the sampling rate between asynchronous microphones. However, since the synchronization accuracy varies with the source signal, noise reduction performance is often degraded. To achieve speech enhancement without accurate synchronization adjustment, we applied Wiener filtering to the highest signal-to-noise ratio (SNR) microphone observation signal. To design a Wiener filter, which is composed of the estimated power spectrum density (PSD) of a target source and other noises, it is necessary to estimate the sensitivities between sound sources and microphones and the number of sound sources without prior information. Under the assumption that source signals are sparse in the time-frequency domain, the sensitivities from a sound source to the microphones are estimated from observed signals. By classifying the estimated sensitivities, the number of sound sources is estimated. Through experiments with asynchronous microphones implemented in smart-phones, we confirmed that the sensitivities between sound sources and microphones were estimated accurately and that the output SNR was improved with the proposed method.