基于EMD的语音信号压缩感知算法

Speech signal compressed sensing algorithm based on EMD

文中将压缩感知理论和经验模态分解方法(empirical mode decomposition,EMD)相结合,用于语音信号压缩上,提出了一种基于EMD的语音信号压缩感知算法。首先用EMD将语音信号分解,可得语音信号的本征模函数信号分量。然后仿真实验模拟EMD分解的过程,并验证本征模函数信号分量的稀疏性。最后结合压缩感知理论基础分别对各个信号分量进行观测抽样,以实现语音信号的压缩。由仿真实验结果可知,语音信号经EMD分解后得到的信号分量在DCT域上较原始语音信号有更好的稀疏性,并且将该算法压缩重构还原出的信号与常规的基于DCT的压缩感知算法以及基于近似KLT的压缩感知算法相比较有更高的平均信噪比,重构性能更佳。

A compressed sensing algorithm based on empirical mode decomposition（ EMD） for speech signal is presented by combining the compressed sensing theory with the empirical mode decomposition method. Firstly,the date by EMD method are processed to obtain the intrinsic mode function components.Then,its better sparse is verfied than the original signal. Finally,each sampled signal component is compressed. Simulation results show that the intrinsic mode function components have better sparse than the original speech signal in DCT domain,and the reconstructed signal has relatively high segmental signal to noise ratio（ SNR） than the original speech signal compressed sensing reconstruction.