论文部分内容阅读
本文提出了一种由连续隐马尔可夫模型（CDHMM）与多层感知器（MLP）构成的混合模型，并将该模型应用于语音孤立词识别。这种混合模型首先用CDHMM来获取输入信号的动态特性，然后再以MLP分类器对输入信号进行分类识别。其主要目的是通过MLP分类器，对CDHMM中的似然估计值进行分析、分类，以加强和提高CDHMM的分类能力。根据这种混合模型，我们建立了一个含30个英语单词的语音识别系统。实验结果表明，该系统的识别率明显高于传统的CDHMM方法。
This paper presents a hybrid model consisting of Continuous Hidden Markov Model (CDHMM) and Multilayer Perceptron (MLP), and applies the model to speech isolated word recognition. This hybrid model first uses CDHMM to obtain the dynamic characteristics of the input signal, and then use the MLP classifier to classify the input signal. Its main purpose is to analyze and classify the likelihood estimates in CDHMM through MLP classifier so as to strengthen and improve the classification ability of CDHMM. Based on this hybrid model, we built a speech recognition system with 30 English words. Experimental results show that the recognition rate of the system is significantly higher than the traditional CDHMM method.