论文部分内容阅读
提出一种基于统计声学模型的单元挑选语音合成算法.在模型训练阶段,首先提取语料库中语音数据的频谱、基频等声学参数,结合语料库中的音段和韵律标注来估计各上下文相关音素对应的统计声学模型,使用的模型结构为隐马尔柯夫模型.在合成阶段,以使目标合成句对应的声学模型具有最大的似然值输出为准则,来进行最佳合成单元的挑选,最后通过平滑连接各备选单元波形来生成合成语音.以此算法为基础,构建一个以声韵母为基本拼接单元的中文语音合成系统,并通过测听实验证明此算法相对传统算法在提高合成语音自然度上的有效性.
This paper proposes a unitized speech synthesis algorithm based on statistical acoustic model.Firstly, the acoustic parameters such as frequency spectrum and fundamental frequency of the speech data in the corpus are extracted and the corpora and the corpus are used to estimate the corresponding phoneme correspondence , And the model structure used is Hidden Markov Model.In the synthesis phase, the best synthesis unit is selected by making the acoustic model corresponding to the target synthetic sentence have the maximum likelihood value output as the criterion, and finally the Based on this algorithm, a Chinese speech synthesis system based on vowels and vowels as the basic splicing unit is constructed, and the experimental results show that this algorithm is more effective than conventional algorithms in improving the naturalness of synthesized speech Validity on.