[Feedback]: Expose POST /v1/audio/transcriptions

[profucius]

Deployment Platform

Docker

Version

1.2.2

Client/Tool

No response

API Endpoint

OpenAI

Issue/Request Description

I want to use this software for processing audio dictation. Currently I am using a software called OpenWhispr which can capture audio by microphone, then pass the audio (completed audio file, not audio stream) to an OpenAI compatible endpoint. The OpenWhispr tries to connect to my AIStudioToAPI instance with the URL I give it, but it returns with an error saying that there is no POST /v1/audio/transcriptions endpoint.

Would it be possible to expose this as an endpoint, so that software that is hardcoded to use the POST /v1/audio/transcriptions endpoint will work as expected? I think the AIStudioToAPI software should be able to handle the data request otherwise.

Logs (Optional)

[INFO] 2026-04-25 08:21:06.041 [America/New_York] [AIStudioToAPI] - [Context#0] Received request: POST /v1/audio/transcriptions
[INFO] 2026-04-25 08:21:07.067 [America/New_York] [AIStudioToAPI] - [Context#0] Request processing failed: Google API returned error: 404 NOT_FOUND {"error":{"code":404,"message":"Requested entity was not found.","status":"NOT_FOUND"}}

[bbbugg]

Thanks for the suggestion.

This endpoint is theoretically feasible by wrapping Gemini's audio input capability into an OpenAI-compatible /v1/audio/transcriptions API.

However, I don't have enough time to implement and maintain this feature recently.

A minimal implementation would need at least:

• parse the OpenAI-compatible multipart/form-data request body
• extract the uploaded audio file from the /v1/audio/transcriptions request
• convert the uploaded audio file into Gemini's audio input format
• convert other OpenAI transcription fields, such as model, language, prompt, response_format, and temperature, into the corresponding Gemini request format where possible
• add a custom transcription system prompt, for example asking Gemini to transcribe the audio completely and only return the transcription text
• support basic json / text response formats
• properly handle file size, MIME type, and errors

Formats like srt, vtt, verbose_json, timestamps, diarization, and streaming would probably be out of scope for an initial implementation, because Gemini's transcription through prompting does not reliably provide OpenAI-compatible timestamps.

I'll mark this as a feature request. PRs are welcome if someone wants to work on a minimal implementation.