Polyphone disambiguation and accent prediction using pre-trained language models in Japanese TTS front-end

TL;DR

This paper introduces a novel approach for Japanese TTS front-end that combines explicit morphological features with implicit features from pre-trained language models to improve polyphone disambiguation and accent prediction, enhancing speech naturalness.

Contribution

It presents a new method integrating PLM-based implicit features with explicit morphological features for better polyphone disambiguation and accent prediction in Japanese TTS systems.

Findings

Improved PD accuracy by 5.7 points

Enhanced AP accuracy by 6.0 points

TTS naturalness close to ground-truth pronunciation

Abstract

Although end-to-end text-to-speech (TTS) models can generate natural speech, challenges still remain when it comes to estimating sentence-level phonetic and prosodic information from raw text in Japanese TTS systems. In this paper, we propose a method for polyphone disambiguation (PD) and accent prediction (AP). The proposed method incorporates explicit features extracted from morphological analysis and implicit features extracted from pre-trained language models (PLMs). We use BERT and Flair embeddings as implicit features and examine how to combine them with explicit features. Our objective evaluation results showed that the proposed method improved the accuracy by 5.7 points in PD and 6.0 points in AP. Moreover, the perceptual listening test results confirmed that a TTS system employing our proposed model as a front-end achieved a mean opinion score close to that of synthesized speech with ground-truth pronunciation and accent in terms of naturalness.