英语句子朗读发音中的音素自动检错系统  

PHONEME ERRORS AUTOMATIC DETECTION SYSTEM FOR THE PRONUNCIATION OF ENGLISH SENTENCES READING

作　　者：丁晓飞[1] 胡国平[1] 戴礼荣[1] 

机构地区：[1]中国科学技术大学讯飞语音实验室,安徽合肥230027

出　　处：《计算机应用与软件》2009年第9期149-152,共4页Computer Applications and Software

摘　　要：为探索智能语音技术在英语发音学习中的作用,开展了面向中国人朗读英语句子的音素发音自动检错技术研究。首先收集了45个人录制的900句英文朗读发音,并由两位专家对音素发音中的错误进行详细的标注,然后基于语音识别技术建立的句子朗读发音中音素自动检错系统,并针对中国人英语发音时最为常见的错读和漏读两大问题,分别提出音素独立检错阈值和限定音素对齐识别网络的方法,对音素检错系统进行了优化,显著地提高了系统的性能,最终系统的召回率和正确率分别达到49%和52%,接近人工专家间的69%召回率下59%的正确率的性能。In order to probe into the utilization of intellectualized speech technology in oral-English pronunciation learning, in this paper it studies the phoneme pronunciation errors automatic detection technology for the Chinese in English-reading pronunciation learning. First, there collected 900 sentences of English reading pronunciation which are recorded by 45 different people,and two experts marked the phoneme pronunciation errors detailedly,then a phoneme errors detecting system is constructed for pronunciation sentences reading based on speech recognition technology. Besides, in this paper it proposes the phoneme-dependent thresholds and the restricted phoneme - alignment recognition network to detect the mispronounced and the omitted pronounced phonemes separately, they are frequently occurred in English pronunciation of the Chinese. In this way the phoneme errors detection system is optimized and it makes an effective performance improvement to the baseline system. Finally the system＇s recall rate and accuracy rate have reached 49% and 52%, approximating the performance between the two experts which is 59% in accuracy under 69% in recall rate.

关 键 词：语音识别 隐马尔可夫模型 音素检错 限定音素识别网络 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] H311[自动化与计算机技术—计算机科学与技术]