电话频带汉语普通话连续语音识别

Telephone Band Continuous Speech Recognition for Chinese Mandarin

基于帧能量参数和帧与稳态背景噪声状态之间的失真度，将连续的电话频带汉语语音信号分割至用隐马尔可夫链模型表示的以字为单位的语音，采用动态时间规正算法和最小失真度准则，作以字为单位的语音识别，从而实现连续语音识别．实验表明电话频带连续语音识到正确率达到７５％．

Based on the frame energy parameter and the distortion between a frame signal and stable background noise state, an utterance of telephone band Chinese continuous speech signal is divided into character speech that is characterized using Hidden Markov Model (HMM). Using dynamic time warping algorithm and minimizing the distortion, the isolated character speech recognition is performed. Thus continuous speech recognition is realized. The experiments showed that correct rate for the telephone band continuous speech recognition was 75%.