论文部分内容阅读
本文通过对汉语语音特点的分析,结合数字信号处理的有关理论,提出了用1/3音节作匹配基元进行汉语单字音识别的方法。它是按汉语音节结构将单字音分成三个匹配基元进行识别的,然后再把结果相拼组成单字音。这种方法介于音素识别和音节识别之间。小字库(104个音节)实验表明:该方法基本上保持了音素识别法的运算量,存贮量低的优点:同时也保持了音节识别法的识别率较高的优势,是一种值得探索的方法。本文主要介绍该方法的原理及韵母识别部分的实验结果。
Based on the analysis of Chinese phonetic features and the related theories of digital signal processing, this paper proposes a method of using Chinese 1/3 phonetic syllable as matching primitive to recognize Chinese single-word speech. It is based on the structure of Chinese syllable words into three matching primitives for recognition, and then spell the results form a single pronunciation. This method is between phoneme recognition and syllable recognition. Small font (104 syllables) experiments show that this method basically preserves the advantages of the phoneme recognition method with low computational complexity and memory storage, while maintaining the advantage of high syllable recognition rate. It is worth exploring Methods. This paper mainly introduces the principle of the method and experimental results of vowel recognition part.