基于平均模型和误差削减网络的语声转换系统

A voice conversion system based on average model and error reduction network

现阶段用于语声转换的深度学习方法多是通过使用大量的训练数据来生成高质量的语声。该文提出了一种基于平均模型和误差削减网络的语声转换框架,可用于有限数量的训练数据。首先,基于CBHG网络的平均模型使用排除源说话人和目标说话人的多说话人语声数据进行训练;然后,在有限数量的目标语声数据下对平均模型执行自适应训练;最后,提出一种误差削减网络,可以进一步改善转换后语声的质量。实验表明,所提出的语声转换框架可以灵活地处理有限的训练数据,并且在客观和主观评估方面均优于传统框架。

So far,many of the deep learning approaches for voice conversion produce good quality speech by using a large amount of training data.This paper presents an average model and error reduction network-based voice conversion framework that can work with a limited amount of training data.We propose to implement a CBHG based average model that is trained with data from many speakers excluding source and target speakers;then,we propose to perform adaptation with a limited amount of target data;last,we propose an error reduction network that can improve the voice conversion quality even further.The experiments show that the proposed voice conversion framework is flexible to work with limited training data and outperforms the traditional frameworks in both objective and subjective evaluations.