Serving Voice AI at Scale
summary
Claims that state space models replace transformers for real‑time voice by keeping generation O(1) while matching transformer quality, with architectures tuned for controllability like voice cloning, accents and background noise. A deployed stack names Sonic 2 and reports ~40 ms latency for TTS and models that run on edge devices about five times faster than cloud roundtrips. Concrete enterprise use cases: live customer support, real‑time gaming NPCs, and a creator voice marketplace with licensing and monitoring integrations.
AI-written summary · about
chapters (6)
- 0:00Introduction to Cartesia AI
- 0:38Voice AI paradigm shift
- 1:41Engineering for voice AI
- 2:53State space model architecture
- 4:45Q&A and real-world usage
- 15:58Future of interactive AI
related talks
01h 07mBuilding and Scaling an AI Agent Swarm of low latency real time voice bots
22mWhy TTS Models Now Look Like LLMs
26mEnabling the next generation in voice technology
26mInterview
29mAndroid Gets a Voice: Building Real-Time Agents
22mVoice Agents That Think While They Talk
25mEngineering voice agents: Latency, quality, and scale
19mThe Advantages of Building Your AI Stack In house