This AI Is Answering the Phone. Can You Tell?

👉 Sign up for LiveKit Cloud: https://cloud.livekit.io/signup?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=tqLU7gDz7Vo Most voice AI agents sound like robots. Not because of the model, but because of the system prompt. In this video you'll hear a full phone call with an AI agent built on a standard STT-LLM-TTS pipeline. Listen for the filler words, the pauses, the personality. None of that comes from the model. It all comes from how the prompt is written. The techniques shown here include: - Filler words and disfluencies ("um", "so", "ya") - SSML pause tags paired with recovery words - Emotion controls for natural vocal variety - Personality written as audible behaviors, not adjectives Want to build this yourself? The full breakdown is in the article below. 📚 Resources 📚 Article: Prompting Voice Agents to Sound More Realistic - http://blog.livekit.io/prompting-voice-agents-to-sound-more-realistic/ Agent docs: https://docs.livekit.io/agents/?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=tqLU7gDz7Vo 🤝 Join the Community Slack: https://livekit.io/join-slack?utm_source=youtube&utm_medium=video&utm_campaign=devrel&utm_content=tqLU7gDz7Vo #livekit #ai #voiceai