Loading...

Text Independent Speaker Verification Based on Phonetic Evolution

Izadi, Mohammad Rasool | 2014

697 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46479 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ghaemmaghami, Shahrokh; Marvasti, Farrokh
  7. Abstract:
  8. In this thesis, by taking into account the phonetic information contained in the signal spectrum we have been looking for, compact Parameterization and less redundancy, in feature extraction phase. In order to access the consecutive dependency nformation contained in the frame, we’ve put on the agenda, using a linear regression model based on the phonetic structure and developments. Then, to describe the phonetic symbols, it has been considered to the tone modeling and ways of expression and modeling of speaker and background. In fact, we have expressed the speaker and the background as dependent functions to speech phonetics classes. According to phonetic subspaces and time
    speech dependency of speech or a combination of those, these phonetic classes are obtained by classification of similar feature sparse vectors. Finally, with regard to this issue that there is no difference between phonetic classes in the Gaussian mixture model based systems and in addition to the weighting the effective classes by changing the likelihood functions based on the linear regression model, the step has been taken to model the manner of speaking by considering the length of the phonetic class
  9. Keywords:
  10. Linear Regression ; Speaker Verification ; Phonetic Evolutions ; Sparse Subspace Clustering

 Digital Object List