首页> 外文会议>IASTED international conference on artificial intelligence and applications >Data Mining Approach for Prosody Modelling by ANN in Text-to-Speech Synthesis
【24h】

Data Mining Approach for Prosody Modelling by ANN in Text-to-Speech Synthesis

机译:浅谈文字综合效果韵律建模的数据挖掘方法

获取原文

摘要

This contribution describes the artificial neural network (ANN) approach for the modelling of fundamental frequency and a duration of speech unit in a text-to-speech (TTS) synthesis. We try to investigate methods for extracting knowledge from the existing speech databases and minimise the number of optimised neural network parameters to improve the generalisation ability of ANN. We try to especially improve the quality of prosody. The ANN for the modelling of two prosody parameters for a ITS synthesis are trained by natural speech. We applied the GUHA method (General Unary Hypotheses Automaton) [1] for the choice of the most important input parameters, and a standard pruning process of ANN, [9] for optimisation of the generalisation ability.
机译:该贡献描述了用于在文本到语音(TTS)合成中的基本频率和语音单元持续时间的人工神经网络(ANN)方法。我们尝试调查从现有语音数据库中提取知识的方法,并最大限度地减少优化的神经网络参数的数量,以提高ANN的泛化能力。我们试图特别提高韵律的质量。对于其合成的两个韵律参数建模的ANN通过自然语音培训。我们应用了Guha方法(一般Unary Husbotheses Automaton)[1]选择最重要的输入参数,以及ANN,[9]的标准修剪过程,用于优化泛化能力。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号