Design of transfer function and matched filter ratio beamformer from the virtual sound source perspective

Speech enhancement in the presence of multiple speech sources under a reverberant and noisy environment is considered as a difficult yet practical problem in speech signal processing. In this paper, we develop a speech enhancement approach to extract desired speech signal from the corrupt observations. The proposed method uses the transfer function ratio beamformer, the multi-channel adaptive filter algorithm and the matched filter array to perform speech enhancement and dereverberation at two stages. The virtual sound source concept is proposed to simplify the treatment for multiple competing speeches. The competing speeches and stationary noise signals are suppressed at the first stage and the enhanced speech from the first stage which contains reverberation is dereverberated at the second stage. In addition, the transfer function ratio estimation method in a more practical scenario is proposed. The experiments are performed in a real room environment. Objective evaluations under various environmental conditions confirm the superiority of the proposed method and the improvement on the correct rate using Mandarin automatic speech recognition system is also provided.