Could it be possible to apply the same technique to the whisper API? about whisper-timestamped

[Jeronymous]

Hello. Let's clarify first: what do you call the "whisper API"?
OpenAI API? Can you put the link?

[maximerdy1]

Thanks for your responsiveness!
The OpenAI API also offers transcription, using the "large" whisper model, and the costs are way less than the cost of the required CPU, but it doesn't support word-level timestamps yet.
Here is the link: https://platform.openai.com/docs/guides/speech-to-text

[Rayan-Allali]

@maximerdy1 you can use fastapi and make it as api

[Jayavel-S]

Hey! I am also looking for this functionality. I need to utilize the Translations endpoint from the OpenAI's API.

[mkmohangb]

The OpenAI API does support word level granularity, timestamp_granularities=["word"], check the section under Timestamps.

[Jayavel-S]

We did check it. But it isn't for the translation endpoint. The ideal usage would be to have at least segment-level timestamps generated in the translation's endpoint.