Design and Implementation of Wake-on-Voice and Command Recognition Algorithm

Abstract

This paper analyzes methods of speech recognition and lays an emphasize on acoustic model, language model and decoding algorithm based on the output of these two models. Hidden Markov Model is applied to acoustic model to build a state graph of acoustic features of input speech sample. After testing, the arousing rate is above 95%, the mis-arousing-rate is below 5% and response time of the wake-on-voice model is about 0.2s.