We’re reimagining AI voice experiences.

AI voice calling is broken! Agents are painfully slow and difficult to build reliably. Success rates are low, and the experience feels uncanny — even in low-risk scenarios with call durations under two minutes.‍

‍Why? There are too many knobs to tune, and too many conversational elements to handle & test. Developers grind through long cycles of patching, fine-tuning, and adding workarounds — just to make basic conversational elements reliable and sound human. Even then, edge cases still break the experience.

No amount of data will fix the status quo.

Voice AI should be plug-and-play. We need to rethink the entire approach to unlock truly accurate & reliable conversational experiences.

At MetaVoice, we're re-architecting the underlying models (speech-to-text, text-to-speech, and LLMs) from the ground up, towards an experience that is as natural as speaking with a human.