ADAPTIVE SPEECH SIGNALS DEREVERBERATION.

An introduction is made of the use of sequential adaptive filtering to the problem of speech signal dereverberation. Two processing methods are considered. Direct adaptive filtering in the time domain and a new proposition for complex adaptive filtering in separate frequency bands. The underlying algorithm for weights adaptation is the well known LMS algorithm. The input signals are obtained from two microphones and the dereverberation is based on the correlation properties of the speech signal and the reverberation at the two inputs. An analytical model is used to study the performance and properties of the two systems and to choose its parameters. The optimal parameters were found in simulations by means of a weighted spectral distance measure and by informal listening.