Realtime/transcription_sessions API returns 401 even when adding ephemeral key

[kazuya.seki]

thanks, I was able to connect with transcription_session.update with semantic_vad specified.

but it does not seem to behave like semantically judging conversation end, it behaves same as server_vad.
I described the same issue in another topic as well

Semantic VAD might not be working with transcription mode

I’m implementing a realtime API for transcription using Websockets, with Semantic VAD specified, but it seems that with small fillers like ‘umm’ the connection breaks and ‘conversation.item.input_audio_transcription.completed’ is sent. In OpenAI Playground’s real-time API, when I specify semantic VAD with eagerness: low, it waits for about 5 seconds of silence even after saying fillers. Is transcribe mode not working correctly? Or is there something wrong with my current implementation or the w…