基于DCT与神经网络的语音增强方法

本文提出新的语音增强方法。这种新方法先将带噪信号进行离散余弦变换,得到带噪语音信号的DCT系数,并提取这些DCT系数的参数作为神经网络的输入来估计阈值,从而构造阈值函数对变换后的系数进行处理,再进行离散余弦反变换最终达到去噪的目的。本文实验中的噪声采用加性高斯白噪声,仿真实验结果表明新方法具有良好的语音增强效果,特别是在低信噪比情况下,效果更好。

Method of speech enhancement based on DCT and neural network

In this paper a new method of speech enhancement is proposed. After the discrete cosine transfer of the noisy signal, we extract some parameters from these DCT coefficients as the inputs of the neural network in order to get the thresholds. Finally, we operate IDCT on the threshold coefficients to implement speech enhancement. Experiments are carried out in the presence of additive white Gaussian noise. Results show that the new method performs well on speech enhancement, especially on the occasion of low SNR.