藏语安多方言语音合成语料库的设计与构建

Design and Construction of Tibetan Am-do Dialect Speech Synthesis Corpus

语音合成库的构建是研究语音合成技术的数据基础,实现语音合成库设计过程的规范化、具体化和完整化是生成高质量语音库的关键。为了进一步提高藏语语音合成库的质量,文章利用藏文字和藏语安多方言的发音特点,使用最小二乘法对文本语料进行优化筛选后,通过藏语专业主持人发音,建立了60小时的高质量、高自然度、高稳定性的藏语安多方言语音合成库,同时,对照安多发音词典对文本进行安多威利转写以提高语音合成的效果。最后,利用MOS评分标准验证得出语音合成库效果良好,该语音合成库的构建对后续藏语语音合成技术的研究提供了有力的数据支持。

The establishment of speech synthesis database is the basis for researching speech synthesis technology.It is the key to generate high-quality speech database to realize the standardization,specificity and completeness in the design process of the synthesized speech database.In order to further improve the quality of the Tibetan speech database,the pronunciation characteristics of Tibetan characters and Tibetan Am-do dialect are used in this paper,and the text corpus is optimized and screened by the least square method.Then,through the pronunciation of Tibetan professional host,60 hours high quality,high naturalness and high stability speech synthesis speech database of Tibetan Am-do dialect is established.At the same time,Am-do Wylie transcribes the text according to Am-do pronunciation dictionary to improve the effect of speech synthesis.Finally,the Mean Opinion Score(MOS) standard is used to verify that the speech database synthesis effect is good.The established speech database provides strong data support for the subsequent research on Tibetan speech synthesis technology.