Language-Agnostic Meta-Learning for Low-Resource Text-to-Speech with Articulatory Features

TL;DR

This paper introduces a language-agnostic meta-learning approach using articulatory features to enable high-quality text-to-speech synthesis in low-resource languages with minimal data.

Contribution

It proposes a novel method combining articulatory feature embeddings with meta-learning to generalize TTS models to unseen languages and speakers.

Findings

Effective fine-tuning with only 30 minutes of data

Achieved high-quality synthesis in unseen languages

Demonstrated cross-language generalization capabilities

Abstract

While neural text-to-speech systems perform remarkably well in high-resource scenarios, they cannot be applied to the majority of the over 6,000 spoken languages in the world due to a lack of appropriate training data. In this work, we use embeddings derived from articulatory vectors rather than embeddings derived from phoneme identities to learn phoneme representations that hold across languages. In conjunction with language agnostic meta learning, this enables us to fine-tune a high-quality text-to-speech model on just 30 minutes of data in a previously unseen language spoken by a previously unseen speaker.