支持重音合成汉语语音合成系统_朱维彬.pdfVIP

第21 卷　第3 期 中文信息学报 Vol. 21, No. 3 2007 年5 月 JO URNA L OF CHINESE INFO RMATION PROCESSING May, 2007 文章编号:1003-0077(2007)03-0122-07 支持重音合成的汉语语音合成系统 朱维彬 (北京交 大学 信息科学研究所, 北京100044) 摘　要:针对基于单元挑选的汉语语音合成系统中重音预测及实现, 本文采用了知识指导下的数据驱动建模策略。 首先, 采用经过感知结果优化的重音检测器, 实现了语音数据库的自动标注;其次, 利用重音标注数据库, 训练得到 支持重音预测的韵律预测模型;用重音韵律预测模型替代原语音合成系统中的相应模型, 从而构成了支持重音合 成的语音合成系统。实验结果分析表明, 基于感知结果优化的重音检测器的标注结果是可靠的;支持重音的韵律 声学预测模型是合理的;新的合成系统能够合成出带有轻重变化的语音。 关键词:计算机应用;中文信息处理;重音;韵律模型;语音合成 中图分类号:TP391　　　　　文献标识码:A AChinese SpeechSynthesis Systemwith Capability ofAccentRealizing HU We-i bin (Institute of Information Science, Beijing Jiaotong University, Beijing 100044, China) Abstract:To aim to predict and realize Chinese accent in a uni-t selection based speech synthesis system, a dat-a driv- en method w as used to build an accen-t supported prosody module. First, with the help of Accen-t Index detector which had been optimized w ith perceptual annotations, a speech corpus had been auto-annotated with Accen-t Index. Then, a prosody predictive module supporting accent had been trained w ith the corpus. Replaced with the new pro-s ody predictive module, the speech synthesis system could synthesize speech with various levels of accent. The re- sults on the experiments had proved the accuracy of aut-o detected accents, and the validity of the prosody predictor, and also the capability of accent realizing of the speech synthesis system. Keywords:computer application;Chinese information processing;accent;prosodic model;speech synthesis 的韵律模型中,没有关于重音的刻画。 1　引言 实现支持重音的语音合成系统,其核心任务是