Learning to Speak from Text: Zero-Shot Multilingual Text-to-Speech with Unsupervised Text Pretraining

TL;DR

This paper introduces a zero-shot multilingual text-to-speech system that leverages unsupervised text pretraining and cross-lingual transfer to synthesize speech in languages with only textual resources, expanding TTS accessibility.

Contribution

It presents a novel framework combining masked language model pretraining and supervised training with frozen language embeddings for zero-shot multilingual TTS.

Findings

Achieves less than 12% character error rate on unseen languages.

Enables TTS synthesis for low-resource languages using only text data.

Demonstrates high intelligibility in zero-shot multilingual speech synthesis.

Abstract

While neural text-to-speech (TTS) has achieved human-like natural synthetic speech, multilingual TTS systems are limited to resource-rich languages due to the need for paired text and studio-quality audio data. This paper proposes a method for zero-shot multilingual TTS using text-only data for the target language. The use of text-only data allows the development of TTS systems for low-resource languages for which only textual resources are available, making TTS accessible to thousands of languages. Inspired by the strong cross-lingual transferability of multilingual language models, our framework first performs masked language model pretraining with multilingual text-only data. Then we train this model with a paired data in a supervised manner, while freezing a language-aware embedding layer. This allows inference even for languages not included in the paired data but present in the text-only data. Evaluation results demonstrate highly intelligible zero-shot TTS with a character error rate of less than 12% for an unseen language.