XTTS

XTTS-v2 is a voice generation model designed for voice cloning and multilingual text-to-speech tasks. It enables users to clone voices into different languages using a brief 6-second audio sample, eliminating the need for extensive training data. The model supports 17 languages, making it suitable for generating speech in multiple linguistic contexts.

Among its features, XTTS-v2 offers emotion and style transfer capabilities by cloning, allowing for nuanced speech synthesis that can capture different expressive qualities. It also supports cross-language voice cloning, which means a voice can be cloned and rendered in languages other than the one used in the original sample. The model operates at a 24kHz sampling rate, contributing to audio quality. Architectural improvements over the previous version include enhanced speaker conditioning, the ability to use multiple speaker references, and interpolation between speakers. Additional updates focus on stability and improved prosody and audio quality.

XTTS-v2 is available under the coqui-public-model-license. It is identified as the same or similar model that powers Coqui Studio and Coqui API.

As a voice generation model with advanced cloning and multilingual capabilities, XTTS-v2 addresses the need for efficient, expressive, and flexible speech synthesis across languages.