INCREMENTAL ESTIMATION OF REVERBERATIONWITH UNCERTAINTY USING PRIOR KNOWLEDGE OF ROOM ACOUSTICS FOR SPEECH DEREVERBERATION

This paper proposes a new dereverberation method that works with incremental processing. A major problem here is how to estimate parameters of the observation process reliably when only a very short observation is available, for example, at the beginning of each human utterance. For this purpose, the prior knowledge of the room acoustics is incorporated into the proposed method by emplying the prior probability density function representation. The proposed method integrates the prior knowledge and the information obtained from the observed signal based on the Bayes’ rule, and achieves incremental dereverberation effectively.