Cartesia

Cartesia is a real-time AI voice platform whose text-to-speech model, Sonic, is built for live voice agents and applications rather than batch narration. Cartesia's own Sonic page claims first audio in under 90 milliseconds, expressive voices in 44 languages, and delivery that interprets the emotional subtext of a transcript automatically, including non-verbal expressions such as laughter inserted directly into the text. The models are built on State Space Model architecture, which Cartesia credits for its combination of low latency, naturalness, and accuracy on details like order numbers and names. Around the model, Cartesia offers instant voice cloning from about 10 seconds of audio and professional cloning from longer recordings, with verified consent required from the speaker, plus localization that carries emotion, tone, and speaker identity across languages, and custom pronunciation dictionaries for names and domain terms. Deployment is flexible: Sonic can run in Cartesia's cloud, on-premises including air-gapped environments, inside your own VPC on the major clouds, or embedded under an OEM licence. Cartesia lists SOC 2 Type II, HIPAA-eligible deployment with BAAs, and GDPR compliance, with zero data retention available to enterprise customers. Cartesia suits developers and product teams building voice agents, support automation, and spoken interfaces where response latency decides whether the conversation feels human. A free plan exists for evaluation, and paid self-serve plans add commercial usage rights and higher capacity.