Thank you — we built Warble on your model 🐦
Hi Audio8 team,
Just a genuine thank-you. Audio8-TTS-Preview-0.6b is a joy to work with — clean code, lovely quality for its size.
We ended up making a small open release on top of it called Warble: five voices we designed ourselves, packaged to be easy to run, including a 4-bit MLX build that runs on Apple-Silicon phone-class chips. It's Apache-2.0 and credits you front and center — a thank-you, not a pitch, and there's no ask here.
If you're curious: https://huggingface.co/scrappylabsai/warble
Really appreciate you putting this into the world. 🙏
— Brian, ScrappyLabs
Small follow-up to the thanks above — we hit (and fixed) two consumer-Blackwell (sm_120 / RTX 5090) serving snags getting the SGLang adapter's CUDA graphs to capture, plus a couple build gotchas. Wrote them up and offered a PR here in case they're useful upstream: https://github.com/Audio8-AI/Audio8_TTS/issues/8 🙏
Thanks so much, Brian — we really appreciate the kind words!
It’s awesome to see what you’ve built on top of Audio8. Warble looks great—really nice work on the voices, packaging, and the 4-bit MLX support. Thanks as well for making it Apache-2.0 and for the generous attribution.
And thank you for taking the time to track down the Blackwell serving issues and documenting the fixes. I’ve already replied on the GitHub issue, and we really appreciate the detailed report and the upstream contribution.
Thanks again for building with Audio8 and for giving back to the community. Looking forward to seeing what you create next! 🚀
Unfortunately, the quality of the cloned language is still quite different from the original. I have to switch back. Of course, the original version runs a bit slowly.