Acoustic system equalization using channel shortening techniques for speech dereverberation

The use of channel shortening techniques for speech dereverberation is discussed in this paper. This approach is motivated by the observation that early reverberation caused by the early reflections in room acoustics is not perceived as a separate sound to the direct sound but is perceived to reinforce the direct sound and is therefore considered useful with regards to speech intelligibility. Compared with inverse filtering, the convergence rate of iterative channel shortening is much higher, which is significant in real-time speech dereverberation. Two iterative channel shortening techniques are presented in this paper and they are shown to outperform standard inverse filtering approaches in the comparative tests described.