Blind adaptive dereverberation of speech signals using a microphone array

This thesis describes a blind adaptive method for the dereverberation of speech (audio) signals in a closed room environment using multiple microphones, and by using the second-order statistics (correlations) of the reverberated speech signal.; The spatial diversity provided by the microphone array creates the equivalent of multiple channels, where each channel is the impulse response of the source to each microphone and is modeled as an FIR filter. This is equivalent to a single-input multiple-output (SIMO) FIR system. The dereverberation is accomplished through the inversion of the acoustical impulse responses. The inverse filters, i.e. equalizers, are found by minimizing what we refer to as a Reduced Mutually Referenced Equalizers (RMRE) error criterion which finds the equalizers at a subset of all possible delays, making the method applicable to the long nonminimum phase FIR impulse responses that characterize acoustical impulse responses. One advantage of this method is that the equalizers are found directly without the need for an impulse response estimation step. Also, unlike many classical microphone array dereverberation methods based on beamforming, the proposed method does not require the speaker location or the microphones' locations be known or set in predefined positions inside the room.