Prosodic break prediction with RNNs

Abstract

Prosodic breaks prediction from text is a fundamental task to obtain naturalness in text to speech applications. In this work we build a data-driven break predictor out of linguistic features like the Part of Speech (POS) tags and forward-backward word distance to punctuation marks, and to do so we use a basic Recurrent Neural Network (RNN) model to exploit the sequence dependency in decisions. In the experiments we evaluate the performance of a logistic regression model and the recurrent one. The results show that the logistic regression outperforms the baseline (CART) by a 9.5% in the F-score, and the addition of the recurrent layer in the model further improves the predictions of the baseline by an 11%.