Hidden Markov Models (HMM) development for Indonesian-language phoneme on Speech to text transcription system.

Abstract

Speech to text transcription system is a system used to convert a voice signal from a microphone or a telephone into a single or a set of words. Research on speech to text transcription systems has been widely applied. But these systems tend to be developed based on words, hence they are inefficient when developed for a large vocabulary. This study uses Baum Welch algorithm for HMM training, Forward algorithm for HMM testing, and Mel-Frequency cepstral coefficient (MFCC) to extract voice features. Data used in this study consist of 5 words in Indonesian language. Phonemes are generated from the segmentation process, and then trained with Baum Welch algorithm to generate the model. This study produced 10 models. The best accuracy obtained is 82% generated by testing the HMM with 2 States and 5 epochs