Research Of Speech Signal Cooperation Compression Algorithm Based On Compressed Sensing

Speech is the most direct and effective way for human to communicate with each other. In the digital information era, people have higher requirement on speech signal processing and the quality of information, greatly improving information sampling costs.How to compress and reconstruction speech signal with lower sampling rates and fewer samples have great significance. Compressed sensing(CS) based on the sparse of the signal in a transform domain sample and almost accurately recover the signal below the standard of Nyquist sampling theorem. It subvert the traditional methods based on Nyquist.In the development of speech signal processing, there were a series of signal processing methods and algorithms produced, and used widely, such as Fast Fourier Transform(FFT), Linear Prediction(LP), Vector Quantization(VQ) and Empirical Mode Decomposition(EMD) and so on. Combining the foundation of CS with the characteristics of traditional signal technology, research on the compression and reconstruction. In this thesis, the main work and innovation are as follow:(1)We study the fundamental of CS in terms of signal sparse representation,measurement matrix, signal reconstruction algorithms, etc. On the other hand, we analyze speech signal and know about its relevant knowledge, such as its characteristics, some processing technologies and coding schemes and so on.(2)In the linear residual domain, the characteristics of linear prediction based on CS and analytical methods for the speech data are studied. Here we put forward a compressed sensing algorithm based on adaptive distribution of observed points in linear residual domain. The features of can be differentiated by energy among each speech frames. And the observed points are distributed in proportion to occupied energy of each frame speech for overall speech data, improving the efficiency of observed points and the performance of speech signal reconstructed.(3)We put forward a compressed sensing algorithm based on empirical mode decomposition(EMD) for speech signal in, which combining the compressed sensing theory and the empirical mode decomposition method. Process the date by EMD method to get the intrinsic mode functions components. The components can be proved to have better sparse than original signal. We compressed each sampled signal component order to compress original signal with better performance.