Channel and speaker adaptation techniques for robust speech recognition

The severe degradations in speech recognition performance are caused by noise, change of the channels and differences of speakers. The study and improvements for adaptive robust speech recognition methods were performed, which are the adaptation methods via VQ code prototype modification, HMM parameters modification and the canonical correlation based compensation method (CCBC). The experimental results have shown that all of methods can make the speaker-independent recognition system robust to channels and speakers, and the CCBC method is better than others in performance.