利用门控循环编解码网络的语音增强方法

为进一步提高基于深度神经网络的语音增强方法的语音增强性能,针对单独使用循环神经网络对上下文信息利用有限的问题,提出一种利用门控循环编解码网络的语音增强方法。使用编码器解码器架构实现,使用循环神经网络对干净语音特征与含噪语音特征之间的映射关系建模,深入挖掘当前待增强语音帧和上下文信息之间的联系。实验结果表明,所提方法相较于基于深度前馈神经网络的语音增强方法增强语音的质量与可懂度更高,语音增强的性能更好,相较于直接应用门控循环神经网络的语音增强方法,显著提升了增强语音的质量。

Speech enhancement method based on gated recurrent encoder-decoder neural network

To further improve the speech enhancement performance of the speech enhancement method based on deep neural network,a speech enhancement method based on deep gated recurrent encoder-decoder neural network was proposed focusing on the problem that using RNN alone has limited ability to model context information.The deep recurrent neural network was used to learn the mapping relationship between the clean speech and the noises features and correlations between the current frame to be enhanced and the context information.Experimental results show that the proposed method significantly outperforms the method using deep neural network in terms of both speech quality and intelligibility.