ANALYSIS OF DATA BALANCING PROBLEM IN ACOUSTIC MODELING OF AUTOMATIC SPEECH RECOGNITION SYSTEM
Laboratory of Computer Science of the University of Le Mans (LIUM), 72085, France; “STC -Innovation” Ltd., 196084, Saint Petersburg ; PhD student; Researcher
Y. Y. Khohlov
STC Ltd., St. Petersburg; Leading Programmer
Read the full article
Abstract. The problem of data balancing for training of acoustic models for automatic speech recognition system is considered. A metric is proposed which enables an explicit account for the data level in a cluster during triphone clustering. The proposed approach is shown to improve the quality of speech recognition.
Keywords:
automatic speech recognition, GMM-HMM, acoustic modeling, acoustic model training, state tying, data balancing, model clustering, triphones