论文部分内容阅读
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。
A generalized model is proposed to unify dynamic time warping (DTW) technology and hidden Markov model (HMM) into a phonetic acoustic model framework. The analysis shows that the generalized model is closer to the actual situation of speech and has a smaller storage capacity. Compared with discrete HMM and DTW, the experimental results of discrete-HMM and DTW show that the recognition performance of generalized model is quite higher than discrete HMM for DTW, The recognition performance of the model is higher than DTW and discrete HMM.