الفهرس 
Speaker IdentificationOnly 14 pages are availabe for public view
 |  | from | 127 |  |  |
| | | from | 127 | | |
Abstract
This thesis presents a study of speaker recognition rates for speech transmitted through Bluetooth channel as a degraded speech signal, while training phase is made by clean speech. This is based on the Mel-Frequency Cepstral Coefficients (MFCC) for feature extraction from the speech. Different approaches for feature extractions were experimented; features from signal, feature from DCT “Discrete Cosine Transform” of the signal, from signal plus DCT, features from DST “Discrete Sine Transform” of the signal, from signal plus DCT, DWT “Discrete Wavelet Transform” of signal, and from signal plus DWT. The feature extraction from the DCT “Discrete Cosine Transform” of the signal achieves the highest recognition rate. The Neural Network (NN) classification technique is used in the experiment as a classifier. The simulation programming is made by Matlab program. In the training phase of experiment, the MFCCs and polynomial coefficients are estimated to form the feature vectors of the database. In the testing phase, similar features to that used in the training are extracted from the degraded transmitted samples and used for matching.