Cascade vs. speech-to-speech vs. full-duplex voice agents

There's more than one way to build a voice agent, and each one has trade-offs. A cascade chains speech-to-text, an LLM, and text-to-speech, so you can pick the best model for each stage. LiveKit's turn detection knows when you've finished, ignores an "uh-huh", and stops when you cut in. Turn-based speech-to-speech uses one model that hears your tone and replies in its own voice, with less delay. Full-duplex speech-to-speech listens while it talks, so it can backchannel while you speak. The best way to choose is to try them. LiveKit Inference now runs all three with one API key, one bill, and zero data retention. Switching is a small change to your AgentSession, and your agent logic stays the same, in Python and Node.js. Get started: https://cloud.livekit.io/signup?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=BZaWD9QPCME 📚 Resources 📚 Agent docs: https://docs.livekit.io/agents/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=BZaWD9QPCME 🤝 Join the Community: https://community.livekit.io/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=BZaWD9QPCME #livekit #ai #voiceai