07-09-2017, 01:22 PM
Conventional signal processing techniques assume that the signal is stationary and ineffective in recognizing non-stationary signals such as voice signals. Voice signals that are more dynamic can be analyzed with much better accuracy using wavelet transform. The speech recognition system developed is a word-dependent voice verification system that combines RASTA and LPC. The voice signal is filtered using the special purpose speech signal filter using the relative spectral algorithm (RASTA). The signals are desnotized and decomposed to derive the wavelet coefficients and, therefore, a statistical calculation is performed. In addition, the formant or resonance of the speech signal is detected using linear predictive coding (LPC). With the statistical calculation on the coefficients only, the accuracy of the individual voice of the verification sample to its own voice is quite high (about 75% to 80%). The reliability of the signal verification is reinforced by the combination of the implications of these two completely different aspects of the individual voice. For voice comparison purposes, four out of five individuals are verified and the results show a higher percentage of accuracy. System accuracy can be improved by incorporating advanced pattern recognition techniques, such as the Hidden Markov Model (HMM).