基于联合损失函数的语音质量特征增强分析

Speech Quality Feature Enhancement Analysis Based on Joint Loss Function

为确保代价函数能够根据人耳感知特点开展分析过程,设计了一种联合损失函数的语音增强深度学习算法;针对损失函数计算过程加入关于人耳听觉的数据。研究结果表明:本文设计的混合损失函数实现了增强语音质量的明显优化。以常规损失函数MES进行处理时,增强语音发生了明显失真;以联合损失函数进行处理时,可以明显降低增强语音失真程度,获得更优的语音质量。

In order to ensure that the cost function can carry out the analysis process according to the characteristics of human ear perception,a speech enhancement deep learning algorithm combined with loss function is designed on the basis of the above.The auditory data of human ear are added to the calculation process of loss function.The results show that the hybrid loss function designed in this paper can achieve obvious optimization of enhanced speech quality.When the conventional loss function MES is used,the enhanced speech is obviously distorted.When processing with joint loss function,the degree of enhanced speech distortion can be significantly reduced and better speech quality can be obtained.