Development of Filipino Phonetically-balanced Words and phoneme-level Hmms

Abstract: In this paper, we developed two sets of phonetically balanced word (PBW) lists in Filipino and two sets of phoneme-level HMMs (Hidden Markov Model). Two PBW lists were based on textbooks used in public school in Philippines and used to develop speech corpus with the fifty speakers of 25 males and 25 females. In a 2-syllable word list (PBW2), an average accuracy rate of 88.95% for speaker dependent and 82.57% for speaker independent test were achieved. For 3-syllable word list (PBW3), the recognizer achieved an accuracy rate of 90.28% for speaker dependent and 83.30% for speaker independent test.

Keywords: Filipino phonetically balanced words, Filipino word corpus, Hidden Markov Model, Automatic Speech Recognition .