Fact · Languages and accents
5.1times the English error rate
On the FLEURS benchmark, Whisper large-v2's word error rate in Hindi was about 5.1 times its English rate.
Checked against its source on October 17, 2026
- What it measures
- Both rates come from the same model and benchmark, so the ratio shows the gap between the two languages.
- How it is calculated
- 21.5 / 4.2 = 5.1
- What it does not show
- A single model on a single read-speech benchmark. Newer models narrow the gap, so do not treat 5.1 as a general law.