Fact · Languages and accents
The LANGUAGES table in OpenAI's Whisper source code includes Bengali, with the code bn, and Urdu, with the code ur.
Checked against its source on October 16, 2026
In the source's own words
“"th": "thai", "ur": "urdu", "hr": "croatian", "bg": "bulgarian"... "lv": "latvian", "bn": "bengali", "sr": "serbian", "az": "azerbaijani"”
whisper/tokenizer.py: the LANGUAGES table (openai/whisper), OpenAI, GitHub, openai/whisper, 2026, LANGUAGES dictionary.
Open the original- What it measures
- The file lists one code and one name per entry. The two passages quoted are the Urdu entry and the Bengali entry, which sit far apart in the table. The table is the list of languages the package can be asked to transcribe.
- What it does not show
- Being in the table does not show how accurate Whisper is in Bengali or Urdu. The table is a list of accepted languages. Apps built on Whisper may offer only some of them.