Decision tree based segmental duration prediction for Amharic TTS system

This paper reports preliminary results of data-driven decision tree based segmental duration prediction for Amharic Text-to-speech synthesis system. Using a manually annotated speech corpus, we extracted a limited number of important features for the construction of training and testing sets. Then, a decision tree based segmental duration prediction was assessed by objective evaluation of the duration model, using root mean squared prediction error (RMSE) and correlation between actual and predicted durations, and we found promising results. The proposed method can be integrated into the TTS duration module for predicting the duration of speech sounds in various textual, prosodic, and segmental contexts.