论文部分内容阅读
韵律模型一直是语音合成中的研究重点,而重音则是目前韵律研究中的主要难点。在已有的研究工作中重音的定性分析较多,但重音生成则相对较少。该文采用基于重音调整的方法,构建了一个支持重音的隐Markov模型(hid-den Markov model,HMM)语音合成系统。在文本分析模块引入最大熵模型完成了基于文本特征的重音预测,然后根据重音调整韵律参数得到调整后的HMM模型,最后采用基于隐Markov模型的语音合成技术(hidden Markov modelbased speech synthesis,HTS)系统合成语音。实验结果表明:采用该方法能够合成出抑扬顿挫的语音。该方法的优势在于能够灵活地扩展到对其他语音表现力的合成。
Prosodic model has been the focus of research in speech synthesis, while accent is the main difficulty in current prosodic research. There are more qualitative analysis of stress in the existing research work, but the stress generation is relatively less. In this paper, an accented Hidden Markov model (HMM) speech synthesis system is constructed based on the method of stress adjustment. The maximum entropy model is introduced into the text analysis module to complete the textural feature-based stress prediction, and then the prosody parameters are adjusted according to the accents to obtain the adjusted HMM model. Finally, the hidden Markov model based speech synthesis (HTS) system Synthetic speech. The experimental results show that this method can synthesize the cadence voice. The advantage of this approach is the flexibility to extend to other speech-expressive compositions.