Font Size: a A A

Research On Application Of Compressed Sensing Technology In Speech Enhancement And EEG Signal

Posted on:2015-04-18Degree:MasterType:Thesis
Country:ChinaCandidate:P Y WangFull Text:PDF
GTID:2308330461474642Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
Speech signals and EEG (electroencephalogram) signals is an important aspect of digital signal processing, has wide application background. In practice, these signals are often non-stationary and mixed noise signal (usually colored noise). Extraction of pure signal from the noisy signal can improve application effect, not only has the important academic value, but also has applied significance.The main work of this paper consists of two parts. The first part, through researching compressed sensing techniques and speech enhancement algorithm, proposed a new speech enhancement algorithm based on compressed sensing technology. The technique used by this proposed algorithm is using KLT transform matrix as a transform matrix, by analyzing the clean speech signal and noise signal has different sparsity on the KLT (Karhunen-Loeve Transform) matrix. The noisy speech signal is projected on the KLT transform matrix using OMP (Orthogonal Matching Pursuit) algorithm to obtain sparse coefficients, and corrected the sparse coefficients by linear filter, and finally reconstructed the signal. Experimental results show that the proposed algorithm is superior to the classical speech enhancement algorithm at different levels of white noise and colored noise environment.The second part, Studied the EEG signal processing method and the traditional white noise de-noising pretreatment method, the proposed new algorithm is applied to EEG signal de-noising preprocessing, then put forward to improve the algorithm to process EEG signal. Improved Algorithm is used to estimate a the noise correlation matrix estimate from noise previously collected and using it for filtering the EEG signal, and then using the new algorithm proposed for its treatment, and filtering the results by the corresponding inverse filter. Experimental results show that the effectiveness of de-noising and the retention of ERP (event-related potentials) in de-noised EEG signal of the proposed method are superior to the EEG signal processing method based on NLM (Nonlocal means) method.
Keywords/Search Tags:Speech enhancement, Compressed sensing, OMP, EEG signal
PDF Full Text Request
Related items