Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 52620 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Ghaemmaghami, Shahrokh
- Abstract:
- The purpose of this project is to improve current speaker verification techniques with short utterance using phonetic information extraction. I-vector technique is widely used in speaker verification systems. Different speakers span a subspace of universal acoustic space, which is usually modeled by “Universal Background model”. Speaker-specific subspace depends on the voice of speaker. In state-of-the-art speaker verification systems i-vectors are extracted by a factor analysis technique to represent speaker characteristics. Studies demonstrate that voiced phonemes contain more speaker-specific information than unvoiced. In this thesis we have classified voiced frames in order to exploit their power of discrimination and by duplicating them focusing more on voiced frames. In addition, in order to extract more information from limited utterances available we extract pitch of each voiced frame and use it for modeling speakers. The proposed approaches have been evaluated by TIMIT data set. Simulation has been done by MATLAB. With comparison to proposed methods ours have outperformed all the others in term of having the least equal error rate (EER)
- Keywords:
- Speaker Verification ; Identity Vector (I-Vector) ; Text-Independent Speaker Verification ; Phonetic Attributes ; Voiced/Unvoiced Classification
Digital Object List
-
محتواي کتاب
- view
Bookmark