Whisper API Translating Based Off Accent

[fresh steeple]

*Steps to reproduce:

1. Record audio message speaking in a second language with an accent. For example if I am an anglophone speaking French I might say "J'aime les croissants"
2. Send audio message to Whisper API
3. Print text transcription

Expected result:

We expect the printed text to read "J'aime les croissants"

Actual result:

Sometimes the API returns "I like croissants"

Additional information

This seems to particularly impact Turkish speakers practicing their English. So, when a Turkish user speaks English with an accent the API transcribes it in Turkish.

[sharp siren]

Please follow the posting guidelines.

For better assistance from the community or an OpenAI team member, please follow the posting guidelines by clicking the button below.

➜ Need to contact OpenAI support? Visit https://help.openai.com

[glossy summit]

Heyy, I saw this issue in the open source whisper.
For reference: https://github.com/openai/whisper/pull/2004

we can leave this open

[zenith sequoia]

the translation is an togglable setting on whisper API

it is not a bug, it is a feature

the model can be set to always output in some languages