MLS: A Large-Scale Multilingual Dataset for Speech Research

TL;DR

The paper presents MLS, a comprehensive multilingual speech dataset with extensive transcribed audio from LibriVox, enabling advancements in ASR and TTS research across multiple languages.

Contribution

Introduction of MLS, a large-scale multilingual speech dataset with accompanying language models and baseline ASR models for 8 languages.

Findings

44.5K hours of English speech

6K hours of speech across 7 other languages

Baseline ASR models provided for all languages

Abstract

This paper introduces Multilingual LibriSpeech (MLS) dataset, a large multilingual corpus suitable for speech research. The dataset is derived from read audiobooks from LibriVox and consists of 8 languages, including about 44.5K hours of English and a total of about 6K hours for other languages. Additionally, we provide Language Models (LM) and baseline Automatic Speech Recognition (ASR) models and for all the languages in our dataset. We believe such a large transcribed dataset will open new avenues in ASR and Text-To-Speech (TTS) research. The dataset will be made freely available for anyone at http://www.openslr.org.