Thanks — I hear you: you need lower latency and more natural-sounding TTS than Inworld/Deepgram, with quality near or above ElevenLabs and lower cost. Recommendation: • Try Cartesia Sonic 3 via LiveKit Inference. It has Excellent voice quality, ~90ms latency (faster than ElevenLabs’ ~150ms), and is cheaper (about $50/million chars on Build/Ship or ~$37.50/million on Scale). You can switch models/voice in your agent config without extra plugins or API keys. Other options to consider: • Deepgram Aura-2: lower cost (~$27–30/million) and ~100ms latency but voice quality is rated Good (not as natural as ElevenLabs or Cartesia). • Keep Inworld for cheap baseline voice (very low cost) only if you can tolerate lower quality and higher latency (~200ms). Quick next steps I can help with: 1. If you want Cartesia Sonic 3, tell me how you currently configure TTS in LiveKit (agent model name or config file) and whether you use LiveKit Cloud or self-hosted so I can give exact config changes. 2. If you want to audition voices first, I can outline a short test plan (sample texts, latency measurements, and cost estimate) to compare Cartesia vs ElevenLabs vs Deepgram. Which option would you like to try first? Sources: https://livekit.com/pricing.md| | Build voice, video, and physical AI | LiveKit | Build Your First AI Voice Agent in Python: Complete Tutorial | LiveKit | Introducing LiveKit Inference: A unified model interface for voice AI | Liv... | Text-to-speech (TTS) models overview | LiveKit Documentation, with interactive elements