Improved speech processing technique for single channel systems

A single-channel algorithm is proposed for noise reduction in speech signals. The proposed algorithm is based on subspace principles and projects the noisy speech vector onto ?signal? and ?noise? subspaces. The major difference of this algorithm from the traditional spectral subtraction methods, it tries to improve the system making use of one the important feature of the human auditory system which is critical bant principle. According to this approach, in the human auditory system diffrent frequencies recepted in the different points of the basilar membrane and there are some threshold frequency values that abrupt changes in the response occurs in the human ear. Results indicated that the proposed algorithm produced significant improvements compared to traditional methods in stationary noise. Further work is possible to extend the robustness of algorithm to work with nonstationary noise environments. Key Words : Spectral substraction, speech processing, noise reduction