Fact · Speech recognition
44.5thousand hours of English
Multilingual LibriSpeech has about 44.5 thousand hours of English audio, drawn from read audiobooks, and about 6 thousand hours across its other languages.
Checked against its source on October 17, 2026
In the source's own words
“consists of 8 languages, including about 44.5K hours of English and a total of about 6K hours for other languages”
MLS: A Large-Scale Multilingual Dataset for Speech Research, Vineel Pratap et al., arXiv (Computation and Language), 2020, Abstract.
Open the original- What it measures
- MLS is built from the same kind of read audiobook recordings as LibriSpeech, extended to other languages.
- What it does not show
- The hours are approximate. The corpus is read speech from audiobooks, so it inherits the limits of read-speech benchmarks.