FLEURS-R: A Restored Multilingual Speech Corpus for Generation Tasks

TL;DR

FLEURS-R is a newly restored multilingual speech corpus that enhances audio quality across 102 languages, facilitating research in speech generation tasks, especially for low-resource languages, by providing a high-quality dataset for TTS and related applications.

Contribution

This paper introduces FLEURS-R, a high-quality, restored version of the FLEURS corpus, enabling improved speech technology development in low-resource languages.

Findings

Significantly improved speech quality in the restored corpus

Maintained semantic content after speech restoration

Enhanced performance of TTS models trained on FLEURS-R

Abstract

This paper introduces FLEURS-R, a speech restoration applied version of the Few-shot Learning Evaluation of Universal Representations of Speech (FLEURS) corpus. FLEURS-R maintains an N-way parallel speech corpus in 102 languages as FLEURS, with improved audio quality and fidelity by applying the speech restoration model Miipher. The aim of FLEURS-R is to advance speech technology in more languages and catalyze research including text-to-speech (TTS) and other speech generation tasks in low-resource languages. Comprehensive evaluations with the restored speech and TTS baseline models trained from the new corpus show that the new corpus obtained significantly improved speech quality while maintaining the semantic contents of the speech. The corpus is publicly released via Hugging Face.