Phrase break prediction with bidirectional encoder representations in Japanese text-to-speech synthesis

TL;DR

This paper introduces a new phrase break prediction method for Japanese TTS that combines features from a pre-trained language model and BiLSTM, improving accuracy and naturalness of synthesized speech.

Contribution

It integrates implicit BERT features with explicit linguistic features in a BiLSTM framework, enhancing phrase break prediction in Japanese TTS systems.

Findings

3.2 point improvement in F1 score over traditional BiLSTM methods

Achieved a mean opinion score of 4.39 in naturalness, close to ground-truth

Demonstrated effective combination of implicit and explicit features for better prediction

Abstract

We propose a novel phrase break prediction method that combines implicit features extracted from a pre-trained large language model, a.k.a BERT, and explicit features extracted from BiLSTM with linguistic features. In conventional BiLSTM based methods, word representations and/or sentence representations are used as independent components. The proposed method takes account of both representations to extract the latent semantics, which cannot be captured by previous methods. The objective evaluation results show that the proposed method obtains an absolute improvement of 3.2 points for the F1 score compared with BiLSTM-based conventional methods using linguistic features. Moreover, the perceptual listening test results verify that a TTS system that applied our proposed method achieved a mean opinion score of 4.39 in prosody naturalness, which is highly competitive with the score of 4.37 for synthesized speech with ground-truth phrase breaks.