母语与非母语语音识别声学建模  

Native and Non-native Speech Recognition Acoustic Modeling

作　　者：曾定[1] 刘加[1] 

机构地区：[1]清华大学电子工程系清华信息科学与技术国家实验室(筹),北京100084

出　　处：《计算机工程》2010年第8期170-172,共3页Computer Engineering

基　　金：国家自然科学基金委员会与微软亚洲研究院联合基金资助项目(60776800);国家"863"计划基金资助项目(2006AA010101;2007AA04Z223;2008AA02Z414)

摘　　要：为了兼容母语与非母语说话人之间的发音变化,提出一种新的声学模型建模方法。分析中国人受母语影响产生的英语发音变化,利用中国人英语发音数据库自适应得到语音模型,采用声学模型融合技术构建融合2种发音规律的识别模型。实验结果证明,中国人英语发音的语音识别率提高了13.4%,但标准英语的语音识别率仅下降1.1%。In order to tolerance the pronunciation changes between native speakers and non-native speakers,this paper proposes a new modeling method for acoustic model.By analyzing English pronunciation changes caused by Chinese,it uses non-native English pronunciation database to gain the corresponding speech model with adaptive method,and uses acoustic model merging technology to construct a recognition model merged two pronunciation rules.Experimental results show that recognition rate of non-native English increases by 13.4% and recognition rate on native English decreases by 1.1%.

关 键 词：语音识别 非母语 模型融合 

分 类 号：TN912.34[电子电信—通信与信息系统]