Fact · Languages and accents
58percent relative word error rate reduction
Meta's MMS model reduces Whisper's word error rate by a relative 58 percent on the 54 languages that both models support in FLEURS, while supporting over 11 times as many languages.
Checked against its source on October 17, 2026
In the source's own words
“MMS reduces the word error rate of Whisper by a relative 58% while supporting over 11 times the number of languages.”
Scaling Speech Technology to 1,000+ Languages, Vineel Pratap et al., arXiv, 2023, Section 4.
Open the original- What it measures
- The comparison uses FLEURS-54, the 54 FLEURS languages that both Whisper and MMS support. MMS was trained on 44.7K hours of labeled data, compared with 680K hours for Whisper.
- What it does not show
- The comparison is the authors' own. MMS was decoded with n-gram models trained on web data, so the numbers compare models under different decoding setups.