Research On Algorithm Of Speech Enhancement Based On-sub-band Decomposition And Compressive Sensing

In the traditional speech enhancement system, information collection is finished according to the Nyquist sampling theorem. According to this theorem, the sampled data contains all the information of the original signal to accurately reconstruct the original signal, but there is largely redundant. In order to reduce transmission and storage resources required, sampling data will be compressed, while leaving some of the most important information while discarding some unimportant information, which will result in a waste of sampling resources. Compressed sensing as a new theory, not directly in the traditional sense of the data collected, and then compress, but the sampling and compression combined to complete in the process of the acquisition, the sampled data is the number after the compression. Compared with conventional sampling theorem, compressive sensing theory takes only useful information that can greatly reduce the number of samples, it is possible to reduce the cost of sampling, transport and storage. Therefore, the paper studies the algorithm combined compressed sensing with speech enhancement, and proposes a compression sensing speech enhancement algorithm based on sub-band.The whole idea of this algorithm is that the full band speech signal through the analysis filter banks are decomposed into four sub bands of speech signal, followed by the four sub band speech signal as discrete cosine sparse transform into four sub-bands of the sparse signal, again four sub-band sparse signals through Wiener filtering algorithm are filtered, coded, measured, reconstructed into four sub-band reconstructed signals, and finally the four sub-band reconstruction signals are combined and reconstructed into the original speech signal. This paper simulated the compression sensing speech enhancement algorithm based on the full-band speech signal and the sub-band speech signal, and the two were compared. Result shows that compression sensing speech enhancement algorithm based on sub-band has small number of samples, a small amount of calculation, and the running time short advantages.This paper firstly studies the speech enhancement algorithm based on CS. Combined with the mathematical model of CS theory, it describes the overall framework, and it simulates and analyzes the CS algorithm based on speech signal for the sparse representation of the signal, the observation matrix and the reconstruction algorithm in three aspects.Secondly, this paper studies the speech enhancement algorithm with Wiener filter based on sub-band. It introduces the sub band filter, and it simulates and analyzes the speech signal with the Wiener filter based on sub-band, and it studies the effect of sampling rate on the filtering effect.Finally, this paper studies the speech enhancement algorithm based on CS and sub band. It simulates the CS speech enhancement algorithm based on the full band speech signal and sub band speech signal, and the two were compared. Result shows that the CS speech enhancement algorithm based on sub-band has advantage of less running time. And it studies the effect of the noise for the algorithm of the paper from two aspects of the sacristy and the reconstruction error. It studies the wiener filtering algorithm based on the Nyquist sampling, the sampling with CS, the sub-band CS sampling. The result of simulation proves again the advantage of wiener filtering algorithm based on sub-band CS sampling.