an automated speech recognition system for phonological awareness of kindergarten students in filipino

Phonological awareness is the ability to hear and attend to different units (e.g. syllables, onset and rime, and phonemes) of language. It is an essential aspect of a child’s learning development as it is a predictor of a child’s success in reading and spelling. The implemented phonological awareness system includes a Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) based automated speech recognizer with pitch prosody-based data augmentation and vocal tract length normalization for children’s phonological awareness skill training and assessment. This study found that the use of pitch prosody-based augmentation is efficient in addressing the pitch and formant mismatch between the trained adult utterances and tested children’s utterances which entailed a WER of 15.08% or an increase in accuracy by 5.16% in comparison to the baseline system. The vocal tract length normalization (VTLN) adaptation did not show much significant improvement to the augmented system as the accuracy only increased by 0.12% or a WER of 14.96%.