On the use of convolutive nonnegative matrix factorization with mixed penalization for blind speech dereverberation

When a signal is recorded in an enclosed room, it typically gets affected by reverberation. This degradation represents a problem when dealing with audio signals, particularly for applications involving automatic speech and/or speaker recognition. There are some approaches to deal with this issue that are quite satisfactory when multi-channel recordings or learning data are available, but this is not the general case in most human-computer interaction applications, and constructing a method that works well in a general context still poses a significant challenge. In this article, we propose a method based on convolutive nonnegative matrix factorization that mixes two penalizers in order to impose certain characteristics over the time-frequency components of the restored signal and the reverberant components. An algorithm for finding such a solution is described and tested. Comparisons of the results against state of the art methods are presented, showing significant improvement.