Inverse filtering of room acoustics using a nonlinear dynamical approach

Acoustical room reverberations can be removed from speech signals by using inverse filtering, which is basically a blind deconvolution approach. Blind deconvolution usually recovers the original speech signal with the use of statistics. However, research shows that the speech generation process is not truly random. In this paper, we propose an inverse filtering technique called minimum phase space volume (MPSV), which exploits the dynamics rather than the statistics of speech signals. Furthermore, the MPSV technique does not require any statistical information whatsoever about the original speech signal. We compare the performance of the MPSV technique to that of the LS technique in deconvolving a speech signal and find that the MPSV technique consistently outperforms the LS technique