Fact · Languages and accents
53.8percent word error rate
On the Common Voice 9 test set, the Whisper large-v2 model had a word error rate of 53.8 percent for Arabic.
Checked against its source on October 17, 2026
In the source's own words
“Whisper large-v2 53.8 19.9 103.4 14.1 13.5 34.2 14.4 6.4 16.0 9.4 5.6 35.1 39.4”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Table 11 (CommonVoice9), column Arabic.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- Read from the Common Voice table in the appendix. Common Voice is crowd-sourced speech recorded by many volunteers, so it is a different test set from Fleurs.
- What it does not show
- This test set differs from Fleurs in speakers and recording conditions, so the Arabic figures on the two sets should not be subtracted from one another.