基于小波变换和压缩感知的语音信号压缩研究

对语音信号直接进行压缩感知处理，通常压缩的效率不高。针对此问题提出了一种基于压缩感知和小波变换的方法，首先用小波变换的方法对语音信号进行级数分解，然后采用压缩感知的方法对小波低频系数进行压缩，并丢弃高频系数，重构语音信号时高频系数用随机信号来取代。采用此种小波变换的方法，与直接采用压缩感知的方法相比，前者的语音信号MOS值稍有降低，但压缩率比直接压缩感知的方法降低了一倍，说明此方法可大大提高压缩的效率。

Research on Speech Compression based on Wavelet Transform and Compressed Sensing

The compressed sensing technology can compress speech signal, but usually the compression efficiency is lower. This paper proposed a method based on compressed sensing and wavelet transform. First, the speech signal was decomposed by the wavelet transform. Low frequency coefficients of wavelet were compressed by the compressed sensing, and high frequency coefficients were discarded and replaced by random signal in the reconstruction. The speech signal was reconstructed by two methods: wavelet transform plus compressed sensing and direct compressed sensing. Comparing the MOS value of the two reconstructed signals, the MOS value of the first method was a little lower than that of the second, but the compression rate of the first was two times lower than that of the second. Therefore, the combination of wavelet transform and compressed sensing can greatly reduce the compression ratio.