Automatic Prosody Annotation with Pre-Trained Text-Speech Model

TL;DR

This paper introduces a neural model that automatically extracts prosodic boundary labels from text-audio data, improving TTS naturalness and reducing manual annotation efforts.

Contribution

The paper presents a novel pre-trained text-speech neural model that automatically annotates prosodic boundaries, outperforming text-only methods and matching human annotation quality.

Findings

Model outperforms text-only baselines

Automatic annotations are comparable to human labels

TTS trained with model annotations performs slightly better

Abstract

Prosodic boundary plays an important role in text-to-speech synthesis (TTS) in terms of naturalness and readability. However, the acquisition of prosodic boundary labels relies on manual annotation, which is costly and time-consuming. In this paper, we propose to automatically extract prosodic boundary labels from text-audio data via a neural text-speech model with pre-trained audio encoders. This model is pre-trained on text and speech data separately and jointly fine-tuned on TTS data in a triplet format: {speech, text, prosody}. The experimental results on both automatic evaluation and human evaluation demonstrate that: 1) the proposed text-speech prosody annotation framework significantly outperforms text-only baselines; 2) the quality of automatic prosodic boundary annotations is comparable to human annotations; 3) TTS systems trained with model-annotated boundaries are slightly better than systems that use manual ones.