Post-filtering with multichannel power spectral estimation using joint diagonalization in multi-speaker environments

This paper addresses the problem of extracting a desired speech signal from the received signals by a microphone array in multi-speaker environments. A new beamformer structure is proposed, which combines a fixed beamformer with a post-filtering technique. In the first stage, a fixed beamformer is designed to spatially extract the desired speech signal by suppressing other undesired speech signals. In the second stage, a post-filter employs a power spectral estimator using the joint diagonalization is proposed to increase the suppression capability. Evaluations using recordings from a real room environment show that the proposed beamformer offers a good interference suppression level whilst maintaining a low distortion level of the desired source.