Dissertation > Excellent graduate degree dissertation topics show

The Underdetermined Blind Source Separation of Speech Signals

Author: YangWen
Tutor: PuJieXin；ZhangHongZuo
School: Henan University of Science and Technology
Course: Applied Computer Technology
Keywords: Underdetermined blind speech separation Independent Component Analysis Sparse Component Analysis Linear membership function K-SCA assumption Hyperplane membership function
CLC: TN912.3
Type: Master's thesis
Year: 2011
Downloads: 44
Quote: 0
Read: Download Dissertation

Abstract


The voice separation as an important research direction of the voice signal processing, speech recognition, speech enhancement has a very positive sense. This thesis is based on analysis and summary of past research, and difficult problems for underdetermined speech separation (traditional algorithms signal sparse separation difficult and separation of noisy low accuracy) study, effective underdetermined Blind Speech Separation applied to noisy, the signal sparse enough scene. The main results are as follows: traditional sparse algorithm separation underdetermined problem of low precision circumstances, proposed algorithm based on linear membership function. The key step in the speech separation process for the mixing matrix solving. The algorithm for solving the mixing matrix using linear mixed-signal clustering characteristics, mixed-signal vector constructs linear membership function. This function is based on vector angle measuring data belonging to the extent of the function represented by a straight line. Obtain by the function extremum solving, data clustering straight, thus solving the mixing matrix. Finally, the voice signal separation experiments verify that, compared to the potential function method, the average signal-to-noise ratio of the algorithm in the case of underdetermined isolated increase 5db. Traditional algorithm in the problem of low accuracy of the noisy speech separation proposed noise impact factor. Based on the impact of noise on the signal data, the influence of noise factor concept, to distinguish the data. Solving the mixing matrix, strengthening the noise impact factor data with low weight, and reduce the influence of noise factor high data weights. Finally, experimental verification, in the case of noisy, combined with noise affecting the signal-to-noise ratio of the signal factor linear membership function method isolated average increase 4db. 3 high shortcomings of the traditional sparse algorithm signal sparse proposed separation algorithm based on the hyperplane membership function owes Blind. 2005 K-SCA assume significantly lower compared with the assumptions of the SCA (sparse component analysis), the sparsity requirements. K-SCA assume that the mixing matrix solving mixed-signal vector points clustering hyperplane normal vector solving. K-SCA assumptions based on hyperplane membership function algorithm. This function is based on the data vector with the function of the angle between the variables measured data vector under the normal vector to the function variable degree of hyperplane through the origin. Through the function extremum solving to get hyperplane data clustering, thus solving the mixing matrix. Finally experiments verify the algorithm can be effectively applied the signal sparse enough in the case, the separated signal has a higher signal-to-noise ratio.

Related Dissertations

  1. The Research of Information Hiding Algorithm Based on Noise Visibility Function and the Independent Component Analysis,TP391.41
  2. Paradigm Design and Algorithm Research for P300-based BCI,TP334.7
  3. Applied Research of Underdetermined Blind Source Separation Method on Phonocardiogram Mixed Signal,TN911.7
  4. Corrosion of the tank bottom Noise in AE Research,TH878
  5. Image feature extraction based image fusion research,TP391.41
  6. Inspection systems for metal abrasive noise cancellation algorithm,TP391.41
  7. Based on odor analysis equipment malfunction detection method,TB17
  8. Telephone-based channel voiceprint recognition algorithm,TN912.34
  9. Design of Visual Experimental Platform Based on Optical Imaging and Microelectrode Array,TP391.41
  10. Multi-Class Pattern Analysis on Human Brain MRI Dataset,TP391.41
  11. Butadiene Rubber Plant and Its Performance Evaluation and Fault Localization Research,TQ333.2
  12. Research on Analysis & Recognition Method for Multi-motor ERD/ERS Signal in BCI System,TN911.6
  13. The Application of Blind Source Separation in the Condition of Completed and Underdetermined in Mechanical Fault Diagnosis,TN911.7
  14. Fault Detection and Diagnosis of Rotating Machinery,TH165.3
  15. Study on Independent Component Analysis-Based Seismic Blind Deconvolution Method and Its Application,P631.4
  16. Research on Blind Separation Methods and Applications of Complex Mixed Acoustic Signals,TN911.7
  17. Independent Component Analysis Appling to Electric Power Harmonic Current Estimation,TM933.1
  18. The Several Trials Extraction of the Sptrague-Dawley Rats’ Flash Visual Evoked Potentials and the Preliminary Research about the Action Potentials,Q42
  19. Measurement of Chlorophyll Content and Distribution in Cucumber Leaves Using Hyper-spectral Imaging Technique,TP391.41
  20. Based on independent component analysis database watermarking technology research,TP309.7
  21. Research of Digital Watermarking Techniques Based on Image and Video,TP309.7

CLC: > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Electro-acoustic technology and speech signal processing > Speech Signal Processing
© 2012 www.DissertationTopic.Net  Mobile