Madurese speech synthesis using HMM

This research is intended to study Madura language which may be the only local language in Indonesia classified into a tonal language. However, the Madurese is not only under- documented in term of phonetics but under-resourced as well. The first step was developing the Madura language voice database. The initial results are limited to the fundamental frequency contour of native male and female utterances. We used a well-known technique called Mel Frequency Cepstral Coefficient (MFCC) to obtain acoustic cues, and the cues were further processed by observing the delta, ? Cepstrum for velocity change in an utterance and the delta-delta ?2, to indicate the acceleration or deceleration of the acoustical cues change over time, respectively. We used instantaneous frequency tracking in STRAIGHT. We obtained results were compared to the well-established TIMIT database in English and Indonesian Speech Database. From the initial results, the ? and ?2 indicated that the Madurese showed a rapid change in both time- and frequency- domain cues. Although the findings may be far from conclusive because the Madura island has four different regions that have its own accent that slightly different from each other. Currently, the on-going research aim is towards a high-quality Madurese speech synthesis.