Research On Reconstruction Algorithm And Denoising Technology Based On Compressed Sensing

Stepping into the information age(big data, intelligence, mobile internet and cloud computing), the link between human becomes closer and the exchange of information becomes more frequent. Traditional digital sampling technology has some disadvantages: Firstly, it needs to sample firstly and compress in the second step, which results in a waste of resources. At the same time, transmitting a large number of data in wireless communications requires a large bandwidth which has brought a big pressure for bandwidth resources. In the past few years, as a new sampling compression technology, compressed sensing brings new vitality to the field of signal processing. The feature of compressed sensing is that it provides both sampling as well as compression simultaneously. In this paper, compressed sensing theory and speech signal processing are combined and this paper mainly research reconstruction algorithms and de-noising methods based on compressed sensing.Firstly, the basic theory of compressed sensing is introduced and the corresponding mathematical model is given. On this basis, the compressive sensing theory and speech signal processing are combined and we analyze the voice compression sensing theory. But in reality, speeches are noisy, so we analyze noisy speech compressive sensing and simulate it.Secondly, the traditional greedy algorithms are described in detail and we analyze the advantages and disadvantages of the greedy algorithms. Aiming at the SAMP algorithm’s disadvantages, VSSAMP is proposed. Through simulation, the proposed algorithm can reconstruct speech signal more accurate and faster.Finally, owing to the observed sequences which are observed by row echelon matrix can retain a large number of the speech characteristics, spectral subtraction is used to denoise. In the side of reconstruction, an adaptive reconstruction algorithm is proposed, which can modify termination parameter as the input signal to noise ratio changed. This algorithm can improve the output signal to noise ratio and reduces the reconfiguration time. Then, a low-pass filter is added to improve intelligibility and naturalness of speech signal. Experimental results show that the speech enhancement algorithm can improve the output SNR and Mean Opinion Score. It has a strong anti-noise capability and can increase the speed of reconstruction.