Voice agents
Voice agents either chain speech-to-text, LLM and text-to-speech, or use a native speech-to-speech realtime model.
Changes fast: check the latest updates too. Reviewed 17 Oct 2026.
Key ideas
- Target under about 800 ms from end of user speech to first audio.
- Key parts: voice activity detection, turn detection, interruption (barge-in) handling.
- Cascaded pipelines give more control and tool reliability; realtime models feel more natural.
- Frameworks: Pipecat, LiveKit Agents; telephony via Twilio or SIP.