Speech Recognition Based on FastIca and MMSE-LSA in Low SNR
LI Shishao;GAO Yong;College of Electronics and Information Engineering,Sichuan University;
Aiming at improving the low speech recognition accuracy in low SNR environment,an algorithm based on FastIca and MMSE-LSA is presented. Firstly,speech is extracted from low SNR signal by FastIca. And then,MMSE-LSA is used to deal with the enhanced speech signal. In this paper,ten numbers from zero to nine are used in isolated words speech recognition. The simulation results show that the speech recognition rate after using the proposed algorithm can more than eighty percent,which improves speech recognition rate in low SNR effectively.