摘要
本文探讨了汉语语音识别的若干问题,并简单介绍了一个大词表汉语语音识别系统,该系统充分考虑了汉语语音的特点,其中主要是汉语语音具有音节性比较强的特点、音节的简单声韵母结构以及汉语以词/词组为语音交流基础的特点.该系统一个显著的特点是系统可以不进行任何训练地添加新词汇,从而使得系统具有比较好的用户接口. 现在系统具有10,000多个词汇,实时测试的平均识别结果是93.1％.
This paper discusses some approaches to Chinese speech recognition andbriefly introduces a practical Large-Vocabulary Chinese Connected speech recognition system. In this system, some characteristics of Chinese speech are taken into consideration, such as Chinese being syllable language, the syllable simply composed of initial consonant, and final vowel, and Chinese words being the basic units in conversation. One of the most important characteristics of this system have a convenient addition of new words, which makes the system have a good user interface. With more than 10,000 words, the system has achieved an average accuracy of 93.1% (top candidate).
出处
《计算机学报》
EI
CSCD
北大核心
1992年第5期364-370,共7页
Chinese Journal of Computers
基金
75重点攻关项目基金
关键词
语音识别
汉语
Syllable segmentation, short-time energy-frequency-value, segmental vector quantization, segmental probabilitical model.
