隐马尔可夫模型和动态时间现正统一框架下的汉语全音节语音识别研究

A study on speech recognition with whole Chinese syllables in unified frame of HMM and DTW

提出了广义模型，将动态时间规正（DTW,DynamicTimeWarping）技术和隐马尔可夫模型（HMM，HiddenMarkovModel）统一到一个语音声学模型的框架内．分析表明，广义模型更接近语音实际情况并具有很小的存储量．还利用广义模型构造了汉语全音节语音识别器，和离散HMM及DTW的对比实验结果显示：对于特定人识别，广义模型的识别性能和DTW相当而高于离散HMM；对于非特定人识别，广义模型的识别性能高于DTW和离散HMM。

In this paper, the Generalized Model (GM) is defined, which unifles Dynamic Time Warping (DTW) and Hidden Markov Model (HMM) illto a frame of acoustical model. Analysis indicates that GMdescribes speech signal more exactly and needs less storage cost. The GM is employed as the recognizer of a speech recognition system of whole Chinese syllables, and the comparative experiment results of GM, Discrete HMM (DHMM) and DTW show that in speaker dependent situation, the performance of GM is equivalent to that of DTW and better than that of DHMM, and in speaker independent situation,is better than that of both DHMM and DTW.