基于小波包变换的多阈值法语音信号去噪净化

文中在小波包变换和传统阈值法的基础上，提出了一种基于小波包变换的多尺度多阈值语音信号去噪净化方法。采用小波包分解，克服了传统的正交小波变换的缺陷。采用多尺度多阈值方法，通过改进噪声方差估计方法，在去噪的同时，进一步提高信噪比。仿真实验结果表明，本方法能有效去除信号中的噪声和较好保留语音细节，达到更佳的语音净化效果。

Multi-threshold Speech Signal Denoising and Purification Based on Wavelet Packet Transform

Through analyzing wavelet packet transform and conventional methods of threshold, a novel improved multi-scale and multi-threshold method based on wavelet packet transform of speech signal denoising and purification is proposed. Wavelet packet decomposition overcomes the weakness of the conventional orthogonal wavelet transform. Multi-scale and multi-threshold method, which amends noise variance estimation, could increase the signal to noise ratio （SNR） while denoises the speech signal. The simulation results show that this novel method could effectively remove the strong noise and retain the principal speech details, thus attaining even better effectiveness of speech signal purification.