HMM-based speech synthesis system with expressive Indonesian speech corpus

In this paper, we present a result of HMM-based speech synthesis system applied to Indonesian expressive speech scorpus. The purpose is to observe speech quality of synthesized speech, conversely. Firstly, we selected expressive Indonesian conversation from movie, novel, and drama transcript. We developed speech database based on phonetically balanced sentence set in which consist of 33 Indonesian phonemes and its IPA symbols and formed 655 sentences. Three expressive styles were applied, namely happiness, sadness, and anger. We hired four professional theater artist to utter the sentences. Segmentation and labeling was performed by manual to create transcription. Variation is given in kind of expressive style and training data amount. The expressive style-dependent decision trees achieve prosodic conversion. The objective and subjective evaluation process are also analyzed. In objective test is using MCD method earn the best score for happiness expressive style with score 4.2 in 82 training data. Then for sadness with score 5.13 in 81 training data and 5.18 for anger in 80 training data. Subjective test with Mean Opinion Score obtain naturalness for happiness, anger, and sadness with score 3.51, 3.38, and 3,0, respectively. The result shown that quality of the synthetic speech is high in term of naturalness.