人工智能培训

搜索

深度学习论文:用于统计参数语音合成系统的说话者自适应神经声码器(Speaker-adaptive neural vocoders for statistical

[复制链接]
laishi 发表于 5 天前 | 显示全部楼层 |阅读模式
laishi 5 天前 149 0 显示全部楼层
深度学习论文:用于统计参数语音合成系统的说话者自适应神经声码器(Speaker-adaptive neural vocoders for statistical parametric speech  synthesis systems)本文提出了用于统计参数语音合成(SPSS)系统的说话人自适应神经声码器。最近提出的基于WaveNet的神经语音编码系统利用自回归框架成功地生成语音信号的时间序列。然而,为目标说话者构建具有有限训练数据的高质量语音合成系统仍然是一个挑战。为了在有限的训练数据的约束下产生更自然的语音信号,我们采用具有神经声编码模型的有效变化的说话者适应任务。在所提出的方法中,应用aspeaker独立训练方法来捕获嵌入在多个说话者中的通用属性,然后对训练的模型进行微调以表示目标说话者的特定特征。实验结果验证了所提出的具有说话者自适应神经病毒编码器的SPSS系统优于传统的基于源滤波器模型的声码器和具有WaveNet声码器的SPSS系统,这些声码器独立于扬声器独立或训练者。
This paper proposes speaker-adaptive neural vocoders for statisticalparametric speech synthesis (SPSS) systems.Recently proposed WaveNet-basedneural vocoding systems successfully generate a time sequence of speech signalwith an autoregressive framework.However, building high-quality speechsynthesis systems with limited training data for a target speaker remains achallenge.To generate more natural speech signals with the constraint oflimited training data, we employ a speaker adaptation task with an effectivevariation of neural vocoding models.In the proposed method, aspeaker-independent training method is applied to capture universal attributesembedded in multiple speakers, and the trained model is then fine-tuned torepresent the specific characteristics of the target speaker.Experimentalresults verify that the proposed SPSS systems with speaker-adaptive neuralvocoders outperform those with traditional source-filter model-based vocodersand those with WaveNet vocoders, trained either speaker-dependently orspeaker-independently.深度学习论文:用于统计参数语音合成系统的说话者自适应神经声码器(Speaker-adaptive neural vocoders for statistical parametric speech  synthesis systems) R0AJLDCw0dISiZqL.jpg
URL地址:https://arxiv.org/abs/1811.03311     ----pdf下载地址:https://arxiv.org/pdf/1811.03311    ----深度学习论文:用于统计参数语音合成系统的说话者自适应神经声码器(Speaker-adaptive neural vocoders for statistical parametric speech  synthesis systems)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

laishi当前离线
新手上路

查看:149 | 回复:0

快速回复 返回顶部 返回列表