Spaces:
Running on Zero
Apply for a GPU community grant: Academic project
This Space provides a public demo for GibbsTTS, a codec-based zero-shot TTS system based on metric-induced discrete flow matching. The demo allows users to synthesize speech from a short prompt audio and target text.
GPU acceleration is necessary because inference involves a DiT-based codec-token generative model and neural vocoder decoding.
CPU inference is too slow for an interactive demo, as each request takes several minutes to complete.
The Space will be public and used to support reproducibility, research communication, and comparison with other zero-shot TTS systems. We will keep the model and demo optimized for inference only, and the Space can sleep when inactive to reduce resource usage.
Hi @ydqmkkx ,
Thanks for the request. We noticed that another author from the same paper also submitted a request for the same demo.
https://huggingface.co/spaces/caiyiyi1998/gibbstts-demo/discussions/1
Could you coordinate with your co-author and let us know which request should remain active? Once confirmed, we'll assign ZeroGPU to the Space.
Docs for ZeroGPU: https://huggingface.co/docs/hub/spaces-zerogpu
Hi @hysts ,
Thank you for your quick reply.
Sorry for the duplicate requests. We are co-authors of the same paper and were both testing the demo deployment.
We have coordinated, and please keep this request for this Space active. The other request can be closed.
Thank you for considering the ZeroGPU assignment.
Thanks for confirming! I've assigned ZeroGPU to this Space.