| Blind speech signal separation has a wide range of potential applications in our life, such as speech enhancement for speech recognition, teleconference application, hearing aids etc. The ultimate aim of blind speech signal separation is to mimic the action of a human in a cocktail party situation, where our hearing system can focus on any specific audio source of interest while suppressing all other sources present even in noisy environments. Blind source separation (BSS) provides a good tool to approach this problem. In blind source separation, source signals are estimated using only information observed at receivers and the estimation is performed blindly, without information on source signals and the mixing system.;First we conduct our convolutive speech signal separation in the frequency domain. We propose a convolutive blind signal separation approach for joint speech signal separation and echo cancellation. Then, we suggest a simple means to using the psychoacoustic properties of human auditory system to improve the quality of separated speech signals.;Next, we propose to combine convolutive blind source separation with beamforming to deal with speech separation in heavy reverberation acoustic environment. By exploiting spatial information from beamforming, we maintain the speech separation performance with lower computational complexity.;Because of the inherent problems in frequency domain BSS algorithms, we investigate novel algorithms to improve the convergence and reduce the complexity of time domain convolutive BSS algorithm. We propose the application of MMax partial update algorithm to the time domain convolutive BSS (MMax BSS) to demonstrate that the partial update scheme applied in the MMax LMS algorithm for single channel can be extended to multichannel time domain convolutive BSS with little deterioration in performance and possible computational complexity saving. Also we propose exclusive maximum selective-tap time domain convolutive BSS algorithm (XMax BSS) that reduces the interchannel coherence of the tap-input vectors and improves the conditioning of the autocorrelation matrix resulting in improved convergence rate and reduced misalignment. Moreover, the computational complexity is reduced since only half tap inputs are selected for updating.;In this thesis, we concentrate on issues of relevance to convolutive blind speech signal separation based on the general frame work of Independent Component Analysis (ICA). Both time and frequency domain convolutive BSS algorithms are investigated in our thesis. |
