Currently, I’m using WebRTC on Android to connect to the gpt-realtime-mini model. When the model is playing audio, it’s not very responsive to voice interruptions, and sometimes it can’t be interrupted at all. The expected behavior is that as soon as human speech is detected, the model should stop immediately. How should I configure this?
Are you using semantic VAD?I’m using webrtc in a mobile app and it works well.
This is the doc I used to set it up. You can configure the eagerness to your liking. https://platform.openai.com/docs/guides/realtime-vad
I’m use normal vad. That might be a problem with my physical device. Thanks.