A New Hybrid Post-Filter using a Multichannel Decision-Directed Approach for A Priori SNR Estimation

Abstract:
The multichannel Wiener filter realized by a microphone array-based beamformer combined with a post-filter has been shown to achieve a good noise attenuation performance with good preservation of the quality of the speech component for wideband telephony in cars. However, disturbing musical tones can often still be perceived by using such a post-filter. In this paper, we propose a new hybrid post-filter estimation using a multichannel decision-directed a priori signal-to-noise ratio (SNR) estimation. By employing this newly formed post-filter, we are able to reduce musical tones significantly while achieving an improvement of the noise attenuation performance and maintaining a good quality of the speech component compared to the state-of-the-art post-filter approach.