Speech acquisition in rooms with microphones positioned at a distance from the talker suffers degradation in quality due to reverberation caused by multiple reflections of the sound from surrounding walls and objects. Speech reverberation is therefore an important problem in, for example, hands-free telecommunication applications. One approach to this problem is to perform dereverberation using blind multichannel system identification and inversion. However, most such methods rely on the assumption that the room transfer functions (RTFs) do not share common zeros. In this paper, a two-stage approach is proposed which separately identifies and equalizes the common and the non-common zeros components of the RTFs. Experimental results indicate a considerable improvement using the new method compared to other existing methods.
展开▼