BLIND DECONVOLUTION AND ADAPTIVE ALGORITHMS FOR DE-REVERBERATION

Abstract: De-reverberation of speech signals in a hands-free scenario by adaptive algorithms has been a research topic for several years now. However, it is still a challenging problem because of the nature of common room impulse response (RIR). RIR is generated artificially based on parameters of the room and its intensity depends on the size, shape, dimensions and materials used in the construction of the room. Speech signals recorded with a distant microphone in a usual room contains certain reverberant quality; this often causes severe degradation in automatic speech recognition performance. Resulting, the degradation of speech signal quality leads to reduced intelligibility to listeners. In this thesis Non Blind and Blind Deconvolution algorithms such as Least Mean Square (LMS), Normalized Least Mean Square (NLMS), Recursive Least Square (RLS), Affine Projection Algorithm (APA) & Constant Modulus Algorithm (CMA) are implemented for removing reverberation in a room environment using a single microphone. The performances of these methods are analyzed using Reverberation Index (RR) and Speech Distortion (SD) parameters. The performances of these methods are tested for two different room sizes and three different reflection coefficients with a total of six different setups for five different filter orders.