Tree-based modeling of prosodic phrasing and segmental duration for Korean TTS systems

Abstract

This study describes the tree-based modeling of prosodic phrasing, pause duration between phrases and segmental duration for Korean TTS systems. We collected 400 sentences from various genres and built a corresponding speech corpus uttered by a professional female announcer. The phonemic and prosodic boundaries were manually marked on the recorded speech, and morphological analysis, grapheme-to-phoneme conversion and syntactic analysis were also done on the text. A decision tree and regression trees were trained on 240 sentences (of approximately 20 min length), and tested on 160 sentences (of approximately 13 min length). Features for modeling prosody are proposed, and their effectiveness is measured by interpreting the resulting trees. The misclassification rate of the decision tree was 14.46%, the RMSEs of the regression trees, which predict pause duration and segmental duration, were 132 and 22 ms, respectively, for the test set. To understand the performance of our approach in the run time of TTS systems, we trained and tested trees with the output of our text analyzer. The misclassification rate and the RMSE were 18.49% and 134 ms, respectively, for prosodic phrasing and pause duration on the test set.