论文部分内容阅读
本文在大量语音实验的基础上,对汉语语音识别方法进行了较为深入的探讨,并以IBMPC/AT配以自行研制开发的TMS320C25-E型高速信号处理板为硬件基础,建立了一个特定人汉语普通话全音节实时识别系统.该系统针对汉语普通话的语音特点,采用了分层识别策略.整个系统响应时间小于0.2秒,用4遍1240个全音节语音对系统进行的严格测试表明:系统四声识别的平均正确率为99%左右,音节识别前5个候选的正确识别率分别为82%,91%,94%,96%,97%;同时,本文根据这一测试结果建立了相应的声韵母混淆矩阵和基于Shepard方法的相似度集群分析树图,并对照汉语语音合成清晰度测试结果及汉语语音知觉结构的集群分析结果,对本系统各部分进行了较为深入的分析,提出了相应的改进措施.
Based on a large number of speech experiments, this paper discusses the Chinese speech recognition method in more depth. Based on the hardware of the TMP320C25-E high-speed signal processing board developed by IBMPC / AT, a Chinese-specific Mandarin syllable real-time identification system.This system adopts the hierarchical identification strategy for the Mandarin Chinese voice.The whole system response time is less than 0.2 seconds, with 4 times 1240 syllables speech rigorous testing of the system shows that: the system four The correct recognition rate is about 99%, and the correct recognition rates of the first five candidates are 82%, 91%, 94%, 96% and 97% respectively. At the same time, this paper establishes the corresponding sound Vowel confusion matrix and similarity cluster analysis tree based on Shepard method. Based on the results of Chinese speech intelligibility analysis and Chinese phonetic perception cluster analysis, the parts of this system are analyzed in depth and corresponding improvements are proposed Measures.