self-host

Miso One is an open-source 8B parameter text-to-speech model with real emotional range and 110ms latency, designed for voiceover work.