User transcription in Realtime in playground is incorrect, althougth the model understood it

API
1

When testing the Realtime recent mode in Playground, the text that is transcribed from my voice input is incorrect, although gpt-4o still understands what I have said.

Why? Do they add in the background a speech-to-text on the user voice input to transcribe it and show in the message history?

2

Yes, exactly. This service (transcription) is decoupled from the actual voice-to-voice service.

1 Like