EMPLOYMENT OF DTW-BASED HMM-GMM MULTI-SESSION TRAINING IN TEXTDEPENDENT SPEAKER VERIFICATION
STC-innovations Ltd., Saint Petersburg, 196084, Russian Federation; ITMO University, Saint Petersburg, 197101, Russian Federation; Senior scientific researcher; Leading engineer
V. A. Sukhmel
St. Petersburg State University, Department of Computer Modelling and Multiprocessor Systems; Post-Graduate Student
A. V. Sholokhov
University of Eastern Finland, Joensuu, FI-80101, Finland; postgraguate
T. . Pekhovsky
«Speech Technology Center-Innovation», Ltd.; leading scientific researcher
Read the full article
Abstract. An HMM training procedure using several password utterances is proposed. The proposed method is based on the Dynamic Time Warping algorithm, and is shown to allow for reduction of verification system errors.
Keywords:
text-dependent speaker verification, short utterance, MFCC, HMM, GMM, DTW
