Development of Filipino Phonetically - Balanced Words and Test Using Hidden Markov Model

In this paper, two sets of phonetically balanced words (PBW) in Filipino were developed; namely the 2syllable, and 3-syllable PBW list. These are tested as a speech corpus in a word-level recognizer using the Hidden Markov Model (HMM) as a framework and Mel-Frequency Cepstral Coefficient (MFCC) as a feature extraction technique. Thus, this study is a preparation for a Largecorpus Filipino Language ASR using HMM. For the testing of the PBW sets, fifty speakers were trained (25 male and 25 female speakers). For the recognition of the 2-syllable word list, an average accuracy rate of 93.25% and 88.67% were achieved for the speaker dependent and speaker independent tests, respectively. For the recognition of the 3syllable word list, the recognizer achieved an accuracy rate of 99.53% and 96.30% for the speaker dependent and speaker independent tests, respectively.