Loading...

Speech Enhancement Based upon Compressed Sensing

Fakhar Firouzeh, Fereshteh | 2014

727 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: English
  3. Document No: 47037 (55)
  4. University: Sharif University of Technology, International Campus, Kish Island
  5. Department: Science and Engineering
  6. Advisor(s): Ghorshi, Alireza
  7. Abstract:
  8. This thesis proposes a novel method for enhancing the speech signal based on compressed sensing. Compressed sensing, as a new rapidly growing research field, promises to effectively recover a sparse signal at the rate of below Nyquist rate. This revolutionary technology strongly relies on the sparsity of the signal and incoherency between sensing basis and representation basis. Exact recovery of a sparse signal will be occurred in a situation that the signal of interest sensed randomly and the measurements are also taken based on sparsity level and log factor of the signal dimension.
    In this research, compressed sensing method is proposed to reconstruct speech signal and for noise reduction. They are formulated in the theoretical framework of compressed sensing using Basis Pursuit (BP) and Compressive Sampling Matching Pursuit (CoSaMP) algorithm when random measurement matrix is utilized to acquire the data. Ultimately, it is demonstrated that the proposed method can perfectly recover the speech signal and compressed sensing-based noise reduction is quite effective in reducing the noise of speech signals, especially for different noises such as white Gaussian noise (WGN), City Rain (CRA), Large Crowd (LCR) and Babble (BAB). Furthermore, the mean opinion score (MOS) of subjective listening quality (LQ) and objective quality score from the perceptual evaluation of speech quality (PESQ) of two different methods (i.e. Basis Pursuit and Compressive Sampling Matching Pursuit) are compared. In addition, the mean opinion score (MOS) of objective quality score of different additive noises are also compared
  9. Keywords:
  10. Compressive Sensing ; Speech Enhancement ; Noise Reduction ; Matching Pursuit ; Basis Pursuit Regularisation (BPR)Algorithm ; Compressive Sampling

 Digital Object List

 Bookmark

No TOC