一种基于语义分析的汉语语音识别纠错方法

汉语语音识别的研究越来越重视与语言处理的结合,语音识别已经不是单纯的语音信号处理。N-gram语言模型应用到语音识别系统中,大大增强了系统的正确率和稳定性,但它也有其自身的局限性,使得语音识别出现许多语法和语义的错误结果。本文分析了语音识别产生语音和文字方面的错误的原因和类型,在概念层次网络语言模型的基础上提出了一种基于语句语义分析和混淆音矩阵的语音识别纠错方法。通过三个发音人、5万字的声音语料和216句实验语句的纠错测试,本文的纠错系统在纠正语义搭配型错误方面有比较好的表现,可克服N-gram语言模型带来的一些缺陷。本文提出的纠错方法还可以融合到语音识别系统中,以便更好地为语音识别的纠错处理服务。

An Error-correct Approach in Chinese Automatic Speech Recognition Based on Semantic Analysis

Now automatic speech recognition (ASR) is not a simplex signal processing. The natural language processing is more and more regarded in Chinese ASR. As a language model, N-gram improved the accurate rate and stability of ASR remarkably. But there are still many syntactic and semantic errors in ASR because of the inherent limitation of N-gram language model. This paper analysed the reson and the types of the phonetic and literal errors in ASR. An error-correct approach in Chinese ASR was proposed in this paper based on sentence semantic analysis, confusion matrix and a language model constructed on hierarchical network of concepts. The error-correct software system runs well especially in correctting the errors of semantic relationship, tested with vocal corpus of 3 person and 50,000 words and with 216 experimental sentences for error-correct. So the new language model constructed on hierarchical network of concepts can overcome the limitation of N-gram model. The approach in this paper also can be merged into ASR to improve the performance of error-correct in ASR.