Malay speaker identification using Neural Networks

This paper investigates the Malay speaker identification using Neural Networks. Speech database was developed with five speakers as trainers and five speakers as imposters. The speech training set included 30 vowel sounds of five trainer speakers. The test set included 30 vowel sounds from the five trainers and 30 vowel sounds from five imposters. The speech sounds were sampled at 20 kHz with 16 bit resolution. A single frame of cepstral coefficients was extracted from the speech sounds using Linear Predictive Coding. Multi-Layer Perceptron with one hidden-layer was used to perform the speaker identification. The output of the MLP consisted of one neuron. Experiments were conducted to determine the optimal signal length of vowels, hidden neuron number and threshold values. A maximum recognition rate of 93.33% was achieved.