Spaces:
Running on Zero
Apply for community grant: Academic project (gpu)
Dear Hugging Face Team,
Our recent project, SongFormer: Scaling Music Structure Analysis with Heterogeneous Supervision, has been submitted to ICASSP 2026, and the arXiv preprint will be available soon.
SongFormer achieves strong performance, and to support the community, we’ve open-sourced our full ecosystem:
- Code: GitHub Repository
- Checkpoints: Hugging Face Model Page
- Training Data (SongFormDB): Dataset on Hugging Face
- Evaluation Benchmark (SongFormBench): Dataset on Hugging Face
We’ve also launched a Hugging Face Space for interactive demos. However, CPU-only inference is too slow for a smooth user experience. We kindly request a free GPU grant for our Space to enable real-time music structure analysis and seamlessly showcase SongFormer’s capabilities.
We believe accessible, high-performance demos are key to advancing music AI research and community engagement.
Thank you for your consideration!
Best regards,
ASLP-lab
Hi @ASLP-lab , we've assigned ZeroGPU to this Space. Please check the compatibility and usage sections of this page so your Space can run on ZeroGPU.
If you can, we ask that you upgrade to Enterprise to enjoy higher ZeroGPU quota and other features like Dev Mode, Private Storage, and more: hf.co/enterprise
Hi @ASLP-lab , we've assigned ZeroGPU to this Space. Please check the compatibility and usage sections of this page so your Space can run on ZeroGPU.
If you can, we ask that you upgrade to Enterprise to enjoy higher ZeroGPU quota and other features like Dev Mode, Private Storage, and more: hf.co/enterprise
They are worthy!
@ASLP-lab Hey guys!
I sorta missed this project of yours so I wanted to stop by and say hi and thank you for publishing it! Its awesome :)
I was wondering if you could advise me on the best way to handle segments boundary detection. Your SongFormer does pretty good job so I was wondering if you can tell me the very basic details?
I am primarily interested in adopting your approach and techniques for instrumental music, specifically MIDI music (symbolic domain).
I tried to estimate boundaries with embeddings similarity drops but it was not very precise. So I was wondering if you can suggest anything here?
Your response and time will be much appreciated!!!
Sincerely,
Alex