Verification of Speech Signal Using Discrete Wavelet Transform and Mel Frequency Cepstral Coefficients Features
摘要
Speaker verification is becoming increasingly important in biometric authentication, especially in systems where both security and user convenience are critical. Traditional methods often struggle to maintain high accuracy under different acoustic conditions. With the aim of closing the existing gaps, the paper presents a better speaker verification algorithm that integrates the Discrete Wavelet Transform (DWT) and the Mel Frequency Cepstral Coefficients (MFCC) to come up with a hybrid algorithm dubbed the W-MFCC. The suggested method would be the use of the DWT to split the speech signal into particular frequency sub-bands, and therein the extraction of the MFCC into a particular frequency band. The process helps the system extract both spectral and temporal characteristics more effectively than the conventional MFCC techniques. The experimental findings are realized as a record of two male speakers pronouncing the word ALLAH on a text-dependent basis, and it is understood that the results give an improvement of approximately 3.6 percent with respect to the traditional system based on MFCC. These results recognize the strengths of the W-MFCC approach, especially in taking into account low-frequency settled details and transient aspects of the speech, and it is therefore a method with potential in the domain of verification of speakers.