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%。

关键词: 语音识别, 非母语, 模型融合

Abstract: 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%.

Key words: speech recognition, non-native, model merging

中图分类号: 

• TN912.34