论文部分内容阅读
虽然隐马尔可夫模型（HMM）是当前最为流行的语音识别模型，但由于一般都采用了状态输出独立假设，因此存在着不能描述语音现象中时间相关性的固有缺陷.本文提出的新模型对语音状态输出特征矢量序列的静态和动态特性信息分别进行参数化建模，然后将它们结合在一起，由此在基于段长分布的HMM（DDBHMM）中引入了帧间相关信息.这种引入帧间相关信息的HMM能够更为精确地描述真实的语音现象.本文在给出新模型的框架后，推导了模型参数的估值公式，并给出了模型的训练和识别算法.汉语非特定人孤立音节的识别实验表明，引入帧间相关信息使HMM的识别性能得到了明显的改善.
Although Hidden Markov Model (HMM) is the most popular speech recognition model at present, it has the inherent flaw that it can not describe the temporal correlation of speech phenomenon because of the independent assumption of state output .However, The speech state outputs the static and the dynamic characteristic information of the characteristic vector sequence to carry on the parametric modeling respectively, then combine them together, thus introduced the interframe-related information in HMM (DDBHMM) based on segment length distribution. The HMM of inter-related information can describe the real speech phenomenon more accurately.After the framework of the new model is given, the evaluation formula of the model parameters is deduced and the training and identification algorithm of the model is given.Non-Chinese The experimental results of isolated syllables show that the introduction of inter-frame correlation information has greatly improved the HMM recognition performance.