Cartesia

Cartesia's pitch is unusually credible for a voice AI startup: its founders literally invented the State Space Model architecture underpinning its Sonic speech generation model, and that research pedigree shows up in genuinely low-latency output that's well suited to real-time conversational agents rather than just pre-rendered narration. Pairing Sonic with Ink for streaming speech-to-text and Line for building full voice agents gives it a more complete stack than point solutions focused only on synthesis, which matters for teams building interactive customer service or healthcare voice products. Deployment flexibility across cloud, on-premise, and on-device is a real differentiator for regulated industries like healthcare and government where data residency matters. The explicit consent-based acceptable use policy around voice cloning is a responsible stance in a space still working out ethical guardrails. Pricing is clear and the free tier's 20,000 monthly credits make evaluation easy, with professional voice cloning gated to the $49/month Startup tier rather than the base Pro plan. Compared to ElevenLabs and other established voice platforms, Cartesia's speed and low-latency focus make it particularly strong for real-time applications, though its voice library and fine-grained expressive control are still catching up to more mature competitors.