LEARNING VECTOR QUANTIZATION FOR ADAPTED GAUSSIAN MIXTURE MODELS IN AUTOMATIC SPEAKER IDENTIFICATION

Speaker Identification (SI) aims at automatically identifying an individual by extracting and processing information from his/her voice. Speaker voice is a robust a biometric modality that has a strong impact in several application areas. In this study, a new combination learning scheme has been proposed based on Gaussian mixture model-universal background model (GMM-UBM) and Learning vector quantization (LVQ) for automatic text-independent speaker identification. Features vectors, constituted by the Mel Frequency Cepstral Coefficients (MFCC) extracted from the speech signal are used to train the New England subset of the TIMIT database. The best results obtained (90% for gender- independent speaker identification, 97 % for male speakers and 93% for female speakers) for test data using 36 MFCC features.