基于HCSIPA的中英文混合语音合成  

Mandarin-English Mixed Text to Speech Based on HCSIPA

作　　者：徐英进[1] 蔡莲红[1] 

机构地区：[1]清华大学计算机科学与技术系,北京100084

出　　处：《计算机工程》2013年第4期14-17,共4页Computer Engineering

基　　金：国家自然科学基金资助项目(60928005,60931160443)

摘　　要：基于双语说话人的中英文混合合成,提出一种中英文通用音标符号——HCSIPA,采用发音方式和部位为构造标准。在中英文混合聚类中,构造针对HCSIPA的中英文共用问题集,以提高中英文在决策树结构上的区分度。实验结果表明,HCSIPA能提高中文和英文的发音单元混合度,减少语言差距带来的音色差距,基于HCSIPA的混合合成系统可以合成较高质量的中英文语音,且中英文混合对单种语言合成的质量下降不明显。A Mandarin-English general phonetic alphabet Human Computer Speech Interaction Phonetic Alphabet（HCSIPA）, in which the articulation manner and place are applied for construction criterion, is proposed in bilingual speaker based Mandarin-English mixed Text to Speech（TTS）. A Mandarin-English common question set suitable for HCSIPA is constructed in Mandarin-English mix clustering. Its construction target is to increase the distinguishing between Mandarin and English in the decision tree structure. The result verifies the HCSIPA can increase the mixing with pronunciation unit of Mandarin and English, to reduce the timbre gap caused by language gap. The mixed TTS based on HCSIPA can synthesize the speech of Mandarin and English with a high quality, and the quality decline to single language synthesis caused by mix of Mandarin and English is not obvious.

关 键 词：HMM混合语音合成 音标符号 HCSIPA符号 双语说话人 音色统一 

分 类 号：TN912.33[电子电信—通信与信息系统]