THE DESIGN OF FEATURE EXTRACTION USING HYBRID MEL FREQUENCY CEPSTRAL COEFFICIENT (MFC) AND PERCEPTUAL LINEAR PREDICTION (PLP) FOR SPEAKER VERIFICATION
|
|
As one of the most important topics in biometrics, speaker recognition is concerned with identifying who spoke an utterance by using the person's voice as a biometric measurement to determine their identity uniquely. Feature extraction as the first stage in the system’s component chain plays a crucial task since the speaker dependent information from speech signal is extracted in this stage. The quality of this stage will strongly affect the quality of next components in the chain. Numbers of methods and techniques have been proposed and most of them are derived from speech recognition techniques like Linear Predictive Coding (LPC), Perceptual Linear Prediction (PLP), Log Area Ratio (LAR), Mel Frequency Cepstral Coefficient (MFCC) etc. Although, there was many improvements to those techniques to fit in speaker recognition purpose, it is still yet to be 100% successful due to the fact that speech is a behavioral biometric that do change over time and susceptible to environment. With the motivation of studying more deeply about feature extraction, it is this research hypothesis that using hybrid techniques in acquiring speech information will improve the uniqueness and quality of speaker information.
|