Audio model

XTTS v2

XTTS v2 is Coqui's open zero-shot text-to-speech model that reproduces a speaker's voice from a six-second reference clip and can read text back in any of 17 languages, including cross-language transfer from a single sample. It supports streaming with latency under 200ms and runs locally on a consumer GPU, and remains widely used for local voice work despite Coqui closing in early 2024.