Introducing BlueTTS
Posted by WeatherZealousideal5@reddit | LocalLLaMA | View on Reddit | 5 comments
I recently worked on BlueTTS, a lightweight text-to-speech model that focuses on speed and usability.
It supports multiple languages: English, Hebrew, Russian, Spanish, and French (even within the same sentence), and comes with a large set of voices available out of the box.
The model reaches up to 1500× real-time on GPU and runs in real-time on CPU, while staying small enough (\~80MB) to run on almost any machine.
Everything is fully open-source, including the training pipeline :)
Contributions are welcome, for example adding support in llama.cpp.
You can check it out here:
geneing@reddit
StyleTTS2 and kokoro are much much better at about the same size.
Elegant-Condition206@reddit
same size yes but faster and open source (has a training code)
Elegant-Condition206@reddit
its a supertonic TTS that was train on libri dataset ( thats why the english is bad..)
Evolution31415@reddit
Hm...
Right from https://github.com/maxmelichov/BlueTTS
AnotherAvery@reddit
The Github page says this projects supports for Hebrew, English, Spanish, Italian and German? Anyways, every Open Source project is appreciated!