一种新的引入MFCC的语音去噪模型

A New Speech Denoising Model with MFCC

为了提高Speech-denoising Wavenet端到端语音去噪模型的去噪效果,将语音的梅尔频率倒谱系数(MFCC)通过一层全连接层和一层卷积层添加至原模型的空洞卷积层之后。实验结果表明,改进后的模型虽然去噪速度降低了18.42%,但是SNR提升了3.60%且训练时间缩短了接近30%。

In order to improve the performance of the end-to-end speech denoising model called Speech-denoising Wavenet,the Mel-frequency cepstral coefficient(MFCC)of the speech is added to the original model after the dilated convolution layer through a fully connected layer and a convolution layer.Experimental results show that although the processing speed is reduced by 18.42%,the signal-to-noise ratio(SNR)is increased by 3.60% and the training time is reduced by nearly 30%.