Advancing Computer Applications

Abstract

Voiceprint identification is an important biometrically based technology for personal identification and verification. The motivation for ende;tvor • stems from the fact that each person has a unique voiceprint pattern, which is formally accepted as a valid personal identification method by law-enforcement agencies. The automation of each system has several facets. In this thesis, the techniques of speech signal pre-processing, feature extraction, and neural networks are combined in order to analyze, identify and classify speech signal. Computer programs are developed to perform the analysis and classification of speech signal. The used speech database consists of240 utterances; 10 repetitions of the same . word, spoken 'by 4 different speakers (2 males, and 2 females). They spoke the Arabic numbers

The used sampling rate was 16,000 sample/sec. i The application of appropriate transforms in feature extraction ' stage (Fast Fourier Transform (FFT), Discrete Hartley Transform

(DHT), Discrete Cosine Transform (DCT), Discrete Walsh-Hadamard Transform (DWHT), Haar Transform (HT), Discrete Wavelet Transform (DWT) ,enabled impotiant information from the speech . signal to be . identified: it's asserted that the use of wavelet transform gives thebest identification ratio (98.75%). The results could possibly be improved further by increasit1g the database and the use of different classification methods (different •neural topology).