Turn Taking
Coming soon. Turn-taking support is in development and will be available in a future release of Orpheus.
What it is
Turn taking decides when one speaker has finished and the other should respond: the core problem in voice agents. It builds on VAD but adds the timing and contextual judgments that separate natural conversation from clipped, talked-over exchanges.
Who it's for
Teams shipping conversational voice AI (agents, copilots, IVR replacements) where end-pointing and response timing materially shape user experience. A voice agent that interrupts users feels broken; one that hesitates too long feels slow. Turn taking is the lever you tune.
What we'll publish here
When Turn Taking ships, this page will cover:
- The Turn Taking output signal and how to consume it in your agent runtime.
- How it interacts with Orpheus enhancement and VAD.
- A complete example matching the format on the Examples page.
- Evaluation guidance. Turn taking is evaluated on end-pointing precision (false starts, missed turns, latency to detected end-of-turn).
If your roadmap needs Turn Taking before the public release, tell us during onboarding.
Voice Activity DetectionSoon
Orpheus Voice Activity Detection (VAD) flags speech vs. silence per chunk in real time for utterance segmentation, silence trimming, and barge-in. Coming soon.
Overview
Integrate Hecttor's Hermes and Orpheus SDKs into any audio pipeline. Learn the standard integration shape plus published recipes for LiveKit and Pipecat.