Text Independent Speaker Recognition Using Mixed MFCC and WOCOR Methods in Persian Language
Hassan Farsi, Saber Amjadi · International Journal of Computer and Electrical Engineering · 2011
Voiced speech is usually used for speaker recognition.But in text-independent speaker recognition it would be better to use special voiced letters which are appeared in all words.In this paper, we have employed the certain letters for speaker recognition.As we know in Persian language, each consonant letter must be followed by a vowel letter.These are A -I -Uæ-eɔ:.Therefore it is enough for text-independent speaker recognition, to find and use these letters.For speaker recognition we employ both vocal source excitation signal and vocal tract system of these letters.Also we use the most prevalent feature parameters for speech/speaker recognition, that is the Mel-frequency cepstral coefficients (MFCC) and for speaker recognition by using vocal source excitation, we have employed Wavelet Octave Coefficients of Residuals (WOCOR).Since these methods are highly sensitive to noise, we use spectral subtraction method to cancel the noise.