Streaming audio and barge-in
Let callers interrupt a prompt and be understood straight away by streaming call audio on Twilio or Infobip.
Normally each turn of a call is a round trip: the flow speaks, the carrier waits for input, and reports it. On some carriers the caller cannot interrupt, because nothing is listening while a prompt plays; on Infobip, for example, keys pressed during a prompt are ignored.
Streaming changes that. The platform holds the call's audio directly, so a caller can speak or press over a prompt and be understood straight away.
What it needs
- An AI voice and a speech-recognition engine in Voice settings. Streaming replaces the carrier's voice and recognition with your own, so both are required. The switch stays disabled until both are set.
- A carrier that supports live audio: Twilio or Infobip. Africa's Talking calls stay on the standard path.
Turning it on
Configure the engines
In Settings › Voice, choose an AI voice engine and a speech-recognition engine, with their API keys.
Switch on streaming
Turn on Streaming audio.
On Infobip, add the endpoint configuration
Enter the WebSocket endpoint configuration ID (below). You can type the id, or pick a stored credential to keep it out of the project.
Infobip: the WebSocket endpoint configuration
Infobip needs a configuration object that holds your project's audio socket. You create it once, against Infobip's API, and paste its id into WebSocket endpoint configuration ID. Without it, calls use the standard path.
The type matters
The configuration must be the WebSocket endpoint type, which joins a call as a participant. Infobip also offers a media-streaming type that only copies a call's audio out: a flow using one connects, carries audio, and is never heard.