FLEURS-R: A Restored Multilingual Speech Corpus for Generation Tasks
This work addresses the need for high-quality speech corpora to advance text-to-speech and other generation tasks in low-resource languages, though it is incremental as it builds on an existing dataset with restoration techniques.
The paper tackled the problem of low-quality speech data for multilingual generation tasks by introducing FLEURS-R, a restored version of the FLEURS corpus with improved audio quality in 102 languages, resulting in significantly enhanced speech quality while preserving semantic content as shown in evaluations.
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.