Miso Labs

Teams building AI voice agents today rely on text-to-speech systems that introduce noticeable lag, with the page citing competitors like Eleven Labs at 700ms and Sesame at 300ms of latency, both slower than the 160ms baseline of human conversational reaction time. This delay creates awkward pauses that break the conversational flow users expect from a voice agent. Additionally, teams that need to keep voice data in-house for sensitive use cases have had to compromise on data sovereignty when using cloud-only TTS providers.

Miso-TTS is a foundation model for voice built around three core capabilities described on the page. It responds in 110ms, faster than typical human reaction time and well ahead of the named competitors, aiming to eliminate the lag that disrupts natural conversation. It also offers one-shot voice cloning, replicating a voice from a ten-second audio sample and maintaining that likeness consistently across a call. Because the models are open source and built for local deployment, enterprise teams can keep voice data on premises, and the company offers on-premises hosting and support contracts on request.