How to initiate realtime transcription session?

[olarcher]

Confirm this is an issue with the Python library and not an underlying OpenAI API

• [x] This is an issue with the Python library

Describe the bug

Support for realtime audio transcriptions was recently announced: https://platform.openai.com/docs/guides/speech-to-text#streaming-the-transcription-of-an-ongoing-audio-recording

I noticed that in the latest release of the python sdk, an AsyncTranscriptionSessions object has become available under client.beta.realtime.transcription_sessions. However, I am not sure how to use this to initiate a new realtime transcription session. Could someone please provide an example?

To Reproduce

Instantiate a new async client with

client = AsyncOpenAI(api_key=OPENAI_API_KEY)

Not sure what to do next to start a realtime transcription session.

Tried:

 async with client.beta.realtime.connect(model="gpt-4o-realtime-preview", extra_query={"intent": "transcription"} ) as conn:
        await client.beta.realtime.transcription_sessions.create(
            input_audio_transcription={
                "model": "gpt-4o-transcribe",
                "language": "de"
            }
        )

        async for message in conn:
                print(message)

Get error: ErrorEvent(error=Error(message='You must not provide a model parameter for transcription sessions.', type='invalid_request_error', code='invalid_model', event_id=None, param=None), event_id='event_BEbGnb8W18CQ9cEZPdORK', type='error')

Code snippets

OS

macOS

Python version

3.12

Library version

openai==1.68.2

[gagan2209]

Same issue with me, have you resolved it ?

[olarcher]

Not yet, falling back to using raw websocket connection for now...

[ANYMS-A]

Met the same issue!

[RobertCraigie]

This is a question for the underlying OpenAI API and not the SDK, so I'm going to go ahead and close this issue.

Would you mind reposting at community.openai.com?

[olarcher]

@RobertCraigie the docs I linked show how this can be achieved with the OpenAI (websocket realtime) API (basically, you have to add an intent query param as follows: wss://api.openai.com/v1/realtime?intent=transcription)

My question is how the same can be achieved with the python SDK. Could you reopen please?