OpenAI Realtime API / transcription : Maximum prompt length of input_audio_transcription?

API
, ,
1

Hello,

I would like to use a long prompt for the transcription in my Realtime API session.

As mentioned here : https://platform.openai.com/docs/guides/realtime-transcription

I use:

input_audio_transcription: {
model: “whisper-1”, // or gpt-4o-mini-transcribe or gpt-4o-transcribe
prompt: transcriptionPrompt,
language: “fr”,
}

But if transcriptionPrompt is longer that 1024 characters, I get a white noise => doesn’t work.

It is written here https://platform.openai.com/docs/models/gpt-4o-transcribe that the context length should be 16 kTokens.

Is it a bug or do I miss something ? If a bug, will the effective context window be increased soon ?

Thanks