الفهرس | Only 14 pages are availabe for public view |
Abstract In this The identification of the speaker is the most important techniques used to identify the speaker through his voice, you can distinguish the speakers through audio characteristics where is extracted audio features of own voice clips of each speaker. In this dissertation, the researchers proposed a model for speaker identification through using Mel frequency Cepstral Coefficients for feature extraction and Deep Neural Networks for the identification stage. The model was implemented and tested against a hybrid of Cochlear Filter Cepstral Coefficients and Deep Neural Networks in three different noise levels 0dB, 6dB, 12dB. The accuracy was evaluated using women utterances only, men utterances only and both. The proposed model proved to have higher accuracy than the CFCC-DNN hybrid in all cases especially in women case at 0dB noise level. Also that women have the highest speaker identification accuracy in all noise levels where, the higher the noise the less the identification accuracy and the higher the number of utterances the higher the identification accuracy. |