Massively Multilingual Speech
Massively Multilingual Speech (MMS)
Speech tech for 1,100+ languages
Latest: MMS 1.0 (May 2023)
Speech-to-text and text-to-speech for 1,107 languages and spoken-language identification for 4,000+ — a 10x leap in coverage, built by pairing wav2vec 2.0 with a surprising data source: recordings of read religious texts (notably the Bible) available in thousands of languages. Halved Whisper's word error rate on its shared languages at the time.
Why it matters
Took open speech recognition from roughly 100 languages to 1,107, with language identification for 4,000+, by pairing wav2vec 2.0 with recordings of read religious texts. It halved Whisper's word error rate on shared languages and made Meta the main supplier of speech tech for low-resource languages.
Facts
- The New Testament exists in audio in over 1,100 languages (~32 hours each) — this pun-free 'found dataset' unlocked the coverage.
- Despite religious-text training, analyses found no measurable theological bias in outputs.
Try it yourself
ASR model on Hugging Face (mms-1b-all) ↗ Run ASR/TTS/LID with Transformers (docs) ↗ TTS model on Hugging Face (English) ↗
Lineage
Sources
arXiv ↗GitHub · fairseq ↗Meta AI blog ↗