摘要： 在汉语TTS系统韵律模型中改善文本处理能力可提高汉语语音合成系统的语音输出质量,针对上述问题,该文提出了竞争型神经网络在汉语TTS(Text to speech)韵律建模中的应用,通过对输入的多个不同韵律特征的模板样本进行竞争,最终选择与自然语音最匹配的那个样本模板,听辨的结果证明,竞争型神经网络模型合成语音的自然度得到进一步的提高。

关键词: 汉语TTS, 韵律模型, 竞争型神经网络

Abstract: A melioration of texts dealing ability can improve naturalness of voice in mandain prosodic model used in TTS system. Based on this,acompeted artificial neural network technology is described and applied to construct the mandain prosodic model in TTS system.Many different prosody-model symbols compete in the input port of the neural network .The system chooses the one which matches the natural voice in the output of the network.They result in noticeably better synthetic speech than the traditional neural-network-based approach, which makes the whole prosodic model more efficientand improves the naturalness of voice.

中图分类号: 

• TP183