【作者】 刘瑞； 飞龙； 高光来； 张红伟；

【Author】 LIU Rui;BAO Feilong;GAO Guanglai;ZHANG Hongwei;College of Computer Science, Inner Mongolia University;

【机构】 内蒙古大学计算机学院；

【摘要】 韵律预测是提高语音合成自然度的重要因素。蒙古语语音合成技术的研究仍处于起步阶段,合成语音的自然度较低,韵律预测成为了蒙古语语音合成亟待解决的关键问题。本文结合蒙古语语音学特点,将蒙古语单词、词性作为特征,采用条件随机场模型对蒙古语韵律短语预测进行了实验比较。实验结果表明,本文所采用方法的韵律短语边界预测F值达到83.28%,为蒙古语韵律预测的进一步研究奠定了基础。更多还原

【Abstract】 Mongolian prosody prediction is a key factor to improve the Naturalness of the Speech and the research of Mongolian speech synthesis technology is still at the initial stage. However, the low naturalness of synthesized speech lead to the prosody prediction has become a key issue to be solved Mongolian speech synthesis. According to the Phonetics Features of Mongolian language. this paper takes the word, part-of-speech as features to do comparison based on conditional random fields. The experimental results have shown that a good performance achieved by the CRF and the prosodic phrase boundary prediction F value reached 83.28%. It laid the foundation for further study of Mongolian prosody phrase prediction.更多还原