A multiple-model speech recognition algorithm is presented in this paper,with which a classifier comprising of a number of Hidden Markov Models(HMMs)is trained for each class in a multiple-class identification problem.The approach employs a training strategy that lays more emphasis on samples that are not correctly identified in earlier attempts.By combining the decisions of the composite HMMs,the multiple-model classifier could well recognize samples with a rather spread-out distribution.The algorithm is applied to an intelligent customer service program to identify phrases and short sentences articulated in different dialects.The experimental results show that the multiple-model classifier outperforms single-model classifier and some speech recognition interfaces in the cases of multiple-dialect and small-vocabulary.
关键词
多模型分类器/隐马尔可夫模型/语音识别/智能客服程序
Key words
multiple-model classifier/hidden Markov model/speech recognition/intelligent customer service program
维普
万方数据