针对藏语的语音合成问题,根据藏语的规律和特点,提出一套完整的基于HMM模型的藏语拉萨语语音合成技术解决方案. 并对其中的关键技术进行阐述,包括合成前端的语料选择、拉丁转写、分词处理、文本分析,以及后端的韵律标注、声码器技术、语音建模、问题集设计等. 实验结果表明,基于该方案搭建的藏语语音合成测试系统有较好的综合得分.%For the issue of Tibetan speech synthesis, we propose a complete set of HMM-base Lhasa Tibetan speech synthesis solutions according to the rules and characteristics of Tibetan.Furthermore we also expound the key technologies in solutions, including corpus selection, representing Tibetan in Latin alphabet, word segmentation and text analysis at the front end of synthesis, and prosodic labelling, the STRAIGHT vocoder, speech modelling, question set designing and so on at the rear end.Experimental results demonstrate that the Tibetan speech synthesis test system constructed based on this set of solutions has a good comprehensive score.
展开▼