When using OpenAI’s Realtime API, how can I make the AI assistant initiate the conversation by speaking first?
I want the user experience to be: click “start conversation” and the AI assistant begins speaking.
I tried sending it a fake user message like, “Hello” to prime the conversation (it has a good intro message if the user actually says hello in the microphone). But I couldn’t get that to work: the voice still doesn’t activate until after the user has spoken and initiated the conversation.
Are you using OpenAI for tts or a 3rd party like Eleven Labs?
There’s no need to pay for a few different audio intros you can simply have pre-recorded and rotate. A thousand button pushes across users to have the same thing output? Even “I’m listening; your command?”.
If you must have AI generation, for example, if there was a user bio offered into the developer message, then you could play audio of your own as input event along the lines, “Please introduce yourself to me!”. Or do that 25% of the time asking for today’s date also, to keep them guessing.
Trigger the “response.create” event after the session is successfully created and the Assistance starts speaking its initial message, ensuring it doesn’t send prematurely when using WebSocket, API, or WebRTC like before session.created event receive.
response_trigger = { "type": "response.create" }
await ws.send(json.dumps(response_trigger)) // for WebRTC or API call adjust it by youself