Text Independent Biometric Speaker Recognition System
Luqman Gbadamosi · International Journal of Research in Computer Science · 2013
Abstract: Designing a machine that mimics the human behavior, particularly with the capability of responding properly to spoken language, has intrigued engineers and scientists for centuries. The earlier research work on voice recognition system which is text-dependent requires that the user must say exactly the same text or passphrase for both enrollment and verification before gaining access. In this method the testing speech is polluted by additive noise at different noise decibel levels to achieve only 75 % recognition rate and would require full cooperation by the speaker which could not be used for forensic investigation. This paper presents the historical background, and technological advances in voice recognition and most importantly the study and implementation of text-independent biometric voice recognition system which could be used for speaker identification with 100% recognition rate. The technique makes it possible to use the speaker's voice to verify their identity and control access to services such as voice dialing, telephone shopping, database access services, information services, voice mail, and remote access to computers. The implementation mainly incorporates Mel frequency Cepstral Coefficient (MFCCs) which was used for feature extraction and Vector quantization using the Linde-Buzo-Gray (VQLBG) algorithm used to minimize the amount of data to be handled. The matching result is given on the basis of minimum distortion distance. The project is coded in MATLAB.