首页> 外文期刊>電子情報通信学会技術研究報告. 音声. Speech >Speaker adaptation using context clustering decision tree for HMM-based speech synthesis
【24h】

Speaker adaptation using context clustering decision tree for HMM-based speech synthesis

机译:基于上下文聚类决策树的说话人自适应,用于基于HMM的语音合成

获取原文
获取原文并翻译 | 示例
       

摘要

In order to synthesize speech with arbitrary individualities and/or emotional expressions, segment-based features have to be used as well as frame-based features. In this paper, to realize MLLR (Maximum Likelihood Liner Regression) based speaker adaptation reflecting those segment-based features for HMM-based speech synthesis, we propose a technique for applying context clustering decision trees constructed in a training stage to tying of regression matrices. Since a set of questions used for constructing context clustering decision trees contains questions related to segment-based features such as position and length, it is possible to incorporate segment-based features into the adaptation. We show that synthesized speech from the adapted model using the proposed technique can have segment-based features.
机译:为了合成具有任意个性和/或情感表达的语音,必须使用基于片段的特征以及基于帧的特征。在本文中,为了实现基于MLLR(最大似然线性回归)的说话人自适应,以反映那些基于片段的特征,用于基于HMM的语音合成,我们提出了一种技术,该技术将在训练阶段构造的上下文聚类决策树应用于回归矩阵的绑定。由于用于构建上下文聚类决策树的一组问题包含与基于片段的特征(例如位置和长度)相关的问题,因此可以将基于片段的特征合并到适应中。我们表明,使用提出的技术从适应模型合成语音可以具有基于片段的特征。

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