基于小波熵自适应阈值的语音信号去噪新方法

针对语音信号去噪问题, 提出小波熵自适应阈值去噪法。首先利用小波变换分解带噪语音信号, 计算小波分解后信号子带区间的小波熵, 然后将小波熵和自适应阈值相结合确定各层高频系数的阈值门限, 采用折中指数阈值函数对各层高频系数进行去噪处理, 重构降噪后的语音信号, 最后对比小波熵自适应阈值、极大极小阈值、固定阈值和无偏风险阈值去噪方法的性能。实验结果表明, 当输入信噪比为5 dB时, 小波熵自适应阈值去噪法的输出信噪比是最大的, 且其输入输出信噪比曲线高于其他三种阈值去噪法的输入输出信噪比曲线, 从而证实该算法具有更好的去噪性能。

New denoising method for speech signal based on wavelet entropy and adaptive threshold

For the problem of speech signal denoising, this paper proposed an adaptive threshold extraction method based on wavelet entropy. The signal with noise was disposed of wavelet decompositon, then calculated the wavelet entropy of the speech signal at decomposed levels. It combined the wavelet entropy and adaptive threshold to decide the threshold of high frequency coefficients. It used compromised index threshold function to denosied speech signal, then reconstructed the speech signals. Compared with the methods of adaptive threshold based on wavelet entropy, max-min threshold, sqtwolog threshold, rigrsure threshold and the proposed threshold. The simulation results show that when the input signal-to-noise ratio (SNR) is 5dB, the output SNR with the method of adaptive threshold based on wavelet entropy is the largest, and the input and output SNR curve is higher than the other three kinds of threshold denoising methods, which proves that this method has better denoising performance.