Hi — first off, thanks for releasing this work, the streaming-buffer formulation is exactly what we needed for a low-latency speech enhancement layer in our voice-assistant pipeline.
The pretrained BBED checkpoint link in the README points to:
https://drive.google.com/drive/folders/1NjNiPj42wZ6NyQT4ykZcwxNjhqkbnLD9?usp=drive_link
Both gdown --folder and a direct browser visit (logged-in and incognito) return HTTP 404 — looks like the folder may have been moved, renamed, or unshared. We're currently falling back to the upstream sp-uhh/sgmse OUVE checkpoint via HuggingFace, which loads cleanly but isn't the BBED variant the paper describes.
Is there a different link that's supposed to be used? If not, could the folder be re-shared with "Anyone with the link", or the checkpoint mirrored to HuggingFace (e.g. under sp-uhh/diffusion-buffer), or file IDs / a direct download URL added to the README?
Happy to help test anywhere that's useful. Thanks again for the work.
Hi — first off, thanks for releasing this work, the streaming-buffer formulation is exactly what we needed for a low-latency speech enhancement layer in our voice-assistant pipeline.
The pretrained BBED checkpoint link in the README points to:
https://drive.google.com/drive/folders/1NjNiPj42wZ6NyQT4ykZcwxNjhqkbnLD9?usp=drive_link
Both
gdown --folderand a direct browser visit (logged-in and incognito) return HTTP 404 — looks like the folder may have been moved, renamed, or unshared. We're currently falling back to the upstreamsp-uhh/sgmseOUVE checkpoint via HuggingFace, which loads cleanly but isn't the BBED variant the paper describes.Is there a different link that's supposed to be used? If not, could the folder be re-shared with "Anyone with the link", or the checkpoint mirrored to HuggingFace (e.g. under
sp-uhh/diffusion-buffer), or file IDs / a direct download URL added to the README?Happy to help test anywhere that's useful. Thanks again for the work.