XTTS

XTTS is the flagship voice cloning model of Coqui TTS, a Python deep learning toolkit for text-to-speech that has been battle-tested in research and production. XTTS v2 clones a voice from a reference clip as short as six seconds, speaks 17 languages with cross-language cloning, and streams audio at 24 kHz with latency under 200 milliseconds, which made it one of the most capable open voice cloning systems of its generation. The wider toolkit spans classic architectures such as Tacotron 2, Glow-TTS, and VITS along with multiple neural vocoders, provides training and fine-tuning recipes for new languages and voices, and through Fairseq integration exposes pretrained models covering more than 1100 languages. The toolkit code is licensed under MPL 2.0, while the XTTS model weights ship under the Coqui Public Model License, which restricts commercial use. Coqui the company shut down in early 2024, but community forks keep the project maintained and its models remain widely deployed.