一种基于马尔可夫模型的汉语语音识别后处理中的音字转换方法

A Kind of Conversion Method from Chinese Spelling to Character Based on Markov Model in the Post Processing of Speech Recognition

为了提高汉语语音识别率，本文根据一种基于马尔可夫模型的统计语言模型去实现汉语音字转换，在实现过程中，提出了它的简化模型，该模型不仅保证了实时性，而且也为以后的工作打下伏笔；同时对训练文本的稀疏问题提出了一种新的解决方案。利用以上模型的模拟实验表明，前向－后向的马尔可夫模型具有较好的识别性能；

In order to improve Chinese speech recognition rate, a kind of conversion method from spelling to character based on Markov Model is employed to realize conversion from spelling to character in this paper. During realization, we put forward a kind of simplified model.The model not only assures the real time characteristics, but also is the base of our future work; at the same time a new solution to the sparseness of training text is put forward. Employed the above model, the simulation test shows that the forward-backward markov model has better recognition characteristics than others. Furthermore, characteristics of models whose output units are words are better than those whose output units are characters.