Fact · Transcription
104.1percent word error rate
On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 104.1 percent for Bengali.
Checked against its source on October 17, 2026
In the source's own words
“Model Afrikaans Amharic Arabic Assamese Azerbaijani Belarusian Bulgarian Bengali Bosnian Catalan Chinese Czech Welsh Danish ... Whisper large-v2 36.7 140.3 16.0 106.2 23.4 45.4 14.6 104.1 15.7”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Table 13 (Fleurs), column Bengali.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- Fleurs is a read-speech benchmark in about 100 languages. Bengali is one of the languages with the highest error rates in the largest Whisper model's table.
- What it does not show
- A word error rate above 100 percent means the output contains more wrong words than the reference has words, which happens when the model inserts words that were not spoken. Read-speech benchmark, from 2022, with the model's own text normalization.