YODAS v3: Over 1 Million Hours of High-Bandwidth, Stereophonic, Multilingual Speech

2026-09-24

AI Summary

This study addresses the scarcity of large-scale, high-fidelity stereo multilingual speech data that constrains model training and evaluation. To this end, it constructs the first weakly annotated 48kHz stereo speech corpus exceeding one million hours across 147 languages. Methodologically, a novel language-balanced collection strategy is proposed, integrating large-scale web crawling, weakly supervised learning, audio quality analysis, and automatic speech recognition (ASR) techniques to ensure data diversity and high quality. The dataset is released under the CC BY 3.0 license, with 22 languages individually exceeding 10,000 hours. Its effectiveness is validated through ASR and neural codec tasks. By providing this critical infrastructure, this work substantially advances multilingual speech research.

Abstract

We present YODAS v3, a weakly-labeled speech corpus containing over 1.1 million hours of 48kHz multi-channel audio in 147 languages, released under a CC BY 3.0 license. YODAS v3 is not only the largest open speech dataset to date, but also the first truly large-scale speech corpus with high-fidelity stereo audio. We first provide the collection methodology for the corpus, where we introduce new techniques for gathering language-balanced speech data. The effectiveness of our approach is shown by the language distribution of the crawled data: 22 languages in YODAS v3 have over 10K hours and 73 languages have over 5K hours of data. We then conduct extensive analyses on the composition of the data, such as the distribution of languages, audio quality, and transcription quality. Finally, we train baseline speech recognition and neural codec models to show the effectiveness of the dataset. Download at https://huggingface.co/datasets/espnet/yodas3.

Problem

• speech corpus
• multilingual
• stereo audio
• large-scale dataset
• weakly-labeled

Innovation

• large-scale speech corpus
• high-fidelity stereo audio
• multilingual
• language-balanced data collection
• weakly-labeled