Word Intelligibility Testing and TTS System Improvement
An approach of word intelligibility evaluation and diagnose of an embedded English TTS system for improvement is presented. The TTS engine is based on the concatenation of variable-length speech units. Multi-Syllable Word Test (MSWT) word corpus is designed for the evaluation. Letter to sound, syllable boundary, prosodic satisfactory, concatenation smoothness, compression distortion as well as the isolated word intelligibility as a whole is tested. Diagnostic analysis cycle is conducted to give the guideline of improving the system. Testing strategy, testing material design, testing tool design and testing procedure are described. Discussion is given. The approach can be applied to different languages and can be adopted for alternate TTS approaches such as HTS.
speech synthesis text-to-speech speech evaluation word intelligibilty
Zhenli Yu Dongjian Yue Yiqing Zu Guilin Chen
School of Engineering and Innovation, Shanghai Institute of Technology, China School of Computer Engineering and Science, Shanghai University, China Anhui USTC iFlyTEK Co, Ltd, China Shanghai Youngtone Technology, China
国际会议
2010 IEEE 10th International Conference on Signal Processing(第十届信号处理国际会议 ICSP 2010)
北京
英文
593-596
2010-08-24(万方平台首次上网日期,不代表论文的发表时间)