vobiz-audio-streams
Installation
SKILL.md
Vobiz Audio Streams skill
Use this for AI voice agents, real-time transcription, custom STT/TTS pipelines, or any workflow that needs raw audio frames rather than just recordings.
How it works
- During a call, your VobizXML answer handler returns a
<Stream bidirectional="true" keepCallAlive="true">wss://your-server/ws</Stream>element. The WebSocket URL is the element's text content, not aurl=""attribute. - Vobiz opens a WebSocket to your server and sends a single
startevent withcallId,streamId,tracks, andmediaFormatnested understart. - Audio frames flow Vobiz → your server as
mediaevents (~50 per second per track, 20 ms each). - For bidirectional streams, your server sends
playAudioevents back, pluscheckpoint,clearAudio, andstopcontrol messages. - When the call ends, the WebSocket closes - there is no inbound
stopevent.
Two ways to start a stream
| Approach | When to use |
|---|---|
<Stream> XML element |
The call is being set up by your answer handler. Best for AI agents that own the call from the first ring. This is the documented path for full-duplex voice agents. |
REST POST .../Stream/ |
The call is already live and you want to attach (or fork) audio mid-call - e.g. start transcription after a transfer, or stream a leg that is in a <Conference>. Returns a stream_id you can later list/get/stop. |