一种新型汉语单音节识别方法

A New Method of Chinese Syllable Vecognition

介绍了一种新型字基 ＶＱ／ＨＭＭ语音识别方法： ＶＱ与 ＨＭＭ分级识别算法。 使得 ＶＱ部分可用作语音识别的第一级处理， ＨＭＭ部分作第二级识别。在第一级识别中 可引进汉语的音素知识，使 ＶＱ／ＨＭＭ性能进一步提高，并可用于大字汇表的实时语音识 别，存贮量、计算量均大大减少。用此方法把汉语四声作一个 ＨＭＭ模型，使汉语 ４００ 个基本音节的识别率达 ９６％以上．若加上精确四声识别则可识别汉语的 １２００种声音。

The paper presents a new word based VQ/HMM speech recognition method. The recognition process is performed hierarchically in two steps. In the preselection step, the words are recognited by VQ method, which includes some phoneme knowledges. A short ranked list of candidates is recognized by the following HMM method. Thus a new VQ/HMM speech recognition system which has less memory, short response time and high performance is developed in the system, four tone speech of one syllable is used to set one HMM model. For the 400 Chinese syllables, the recognition rate is more than 96%. If the four tone is added, the 1200 Chinese basic speech can be recognized with high performance.