Proper Wavelet Identification to Process Vowel Phonemes of Persian Language

To analyze speech signal, wavelet transform is suitable and powerful. Speech signal is non-stationary requiring both time and frequency domains data at processing. By wavelet transform, required data of both domains could be extracted from the signal. This paper studies which wavelet best suits for analyzing vowel phonemes of Persian speech signal by wavelet transform. When a wavelet is found proper to analyze speech signal, output of the method with more suitable wavelet is expected to have more optimized results than the other wavelets in speech processing applications based on wavelet transform.
According to the results, similarity between signals of Persian language vowel phonemes and approximation coefficient of signal analysis was measured by using a special wavelet and finally, by making some calculation, examining correlation coefficient, and comparing results of different wavelets, coif3 & Db5 were found best wavelets to analyze Persian language vowel phonemes. What follows is comparing this method with other methods by studying the effect of using a proper wavelet to decrease noise and audibly improve the signal, and this proves the efficiency of the wavelet used.