Research On Time Delay Estimation For Speech Signal In Room Environment

As an important part of signal processing, Time Delay Estimation (TDE) has beenwidely applied for its simplicity in practice application, especially which is related topositioning, such as localization. However, the complexity of reality environmentchallenges classical methods that based on an ideal model. As TDE usually is the firstpart of signal processing systems, the resolution and robustness of TDE often determinethe ultimate performance of the systems.The speech signal is a common information carrier as well as a typical complicatednon-stationary signal. Reverberation and noise make speech signal processing morecomplex. That’s why indoor speech signal processing as well as indoor TDE has beenthe focus of both academic and industrial long-term concern. This thesis focuses on thistopic and introduces sparse priori to classical TDE methods in case of unknown source.Simulation results show that this kind of priori are indeed able to a certain extent toimprove the performance of the classical methods. The main works and innovations areas follows:1. A novel L1-SVD-based high-resolution TDE algorithm. By establishing ahigh-resolution model for generalized cross-correlation methods and introducing thesparsity of the cross-correlation spectrum of received signal, this algorithm makes useof MUSIC-type frequency estimation and convex programming to achieve highresolution and robust to noise is proposed.2. A sparse-LMS-based adaptive TDE algorithm is proposed. Based on a classicaladaptive TDE method, this algorithm exploits the sparsity of weighting function toincrease the performance and convergence of adaptive TDE methods.3. A sparse Blind-Channel-Identification (BCI) TDE algorithm is proposed. Basedon a cross relation for BCI methods, this algorithm exploits the sparsity of channels toenhance BCI’s robust to noise and error of channel length estimation.