OmniVoice free TTS model supports more than 600 languages
The k2-fsa project's OmniVoice clones a voice from short audio samples and offers voice design by gender, age and pitch, according to Gigazine.
OmniVoice, a free text-to-speech model that supports more than 600 languages, has been introduced by the k2-fsa project. The model performs zero-shot voice cloning from short audio samples without any additional training, according to Gigazine.
An official demo page offers samples in 102 languages, including Japanese. The model also supports cross-lingual zero-shot TTS, though an accent from the language of the reference audio can remain in the output.
Voice design and fine control
A Voice Design feature generates speech from a description of gender, age and pitch. Fine-grained control also allows non-verbal sounds such as laughter to be inserted into the text.
OmniVoice can be installed with pip and run through a local web UI on port 8001. The project also provides long-text splitting, saved voice presets and multi-GPU batch processing.
Quick answers
Is OmniVoice free to use?
Yes. OmniVoice is described as a free text-to-speech model introduced by the k2-fsa project.
How many languages does OmniVoice support?
The model supports more than 600 languages, and the official demo page offers samples in 102 languages, including Japanese.
How do I run OmniVoice locally?
It can be installed with pip and run through a local web UI on port 8001.