ultimatevocalremovergui icon indicating copy to clipboard operation
ultimatevocalremovergui copied to clipboard

Does anyone know where i can find the model sami-bytedance?

Open thecatontheceiling opened this issue 1 year ago • 20 comments

I've seen it a couple of times on mvsep but i cant find a clear explanation on what it is and how to get it, any help would be appreacited :D

thecatontheceiling avatar Jun 03 '23 13:06 thecatontheceiling

The SDR scores are off the charts. I cant bother using any other models for my workflow, knowing this model exists. Waiting impatiently :D

Dyslexicon avatar Jun 06 '23 15:06 Dyslexicon

this single model seems to beat full fledged MDX ensembles and htdemucs... it looks good but I can't find it ANYWHERE for the life of me

thecatontheceiling avatar Jun 06 '23 23:06 thecatontheceiling

Have we any stem extraction examples from this model published publicly? You say you saw this model as an option on the mvsep website? https://mvsep.com/ or somewhere else?

Hopefully it is released publicly soon, seems a colossal engineering feat to withhold - after all this was a public competition, seems right and proper to release the results.

Dyslexicon avatar Jun 07 '23 00:06 Dyslexicon

https://mvsep.com/quality_checker/leaderboard2.php?id=2374

I saw it on mvsep here, god the SDR is off the charts lol

thecatontheceiling avatar Jun 07 '23 07:06 thecatontheceiling

https://mvsep.com/quality_checker/leaderboard2.php?id=2374 I'm not sure what this is about, no "Other" stem

https://www.aicrowd.com/challenges/sound-demixing-challenge-2023/problems/music-demixing-track-mdx-23/leaderboards First place winning entry in SDX2023 Competition. On here, it does show results for the "Other" stem, and handily beats all other models.

Dyslexicon avatar Jun 07 '23 13:06 Dyslexicon

@Anjok07 maybe you could convince the makers of this model to bring it to UVR??

Dyslexicon avatar Jun 13 '23 22:06 Dyslexicon

Don't think the model is there.. but here's their page in any case:

  • https://www.ismir2020.net/bytedance/
    • Speech, Audio & Music Intelligence Research

      Welcome to ByteDance booth! We’re SAMI (Speech, Audio & Music Intelligence) team at ByteDance AI Research lab.

There's apparently also an audio separation feature in their new 'Ripple' tool, though unsure if it's the same one as shown on these charts:

  • https://techcrunch.com/2023/06/30/tiktok-parent-bytedance-launches-music-creation-audio-editing-app/

0xdevalias avatar Aug 21 '23 02:08 0xdevalias

A few days ago they released a paper detailing their architecture, and we are one step closer to getting a model that is if not directly from them, recreated using their description for our use.

Zokhoi avatar Sep 13 '23 06:09 Zokhoi

It seems that they used 16 Nvidia V100-32GB GPUs for training. Looking forward to their pre-trained model.

happyTonakai avatar Sep 13 '23 06:09 happyTonakai

https://github.com/lucidrains/BS-RoFormer

There is code, but no pre-trained model.

owlwang avatar Sep 15 '23 10:09 owlwang

Any heroic engineers who can port a functional model of this into Google Colab or UVR, please do! Colab preferable since I dont have a 40GB NVIDIA GPU :)

Dyslexicon avatar Sep 15 '23 23:09 Dyslexicon

Wondering how good sami model is, haven't seen any example result.

assocold avatar Oct 15 '23 07:10 assocold

Still no pretrained model?

zxcvqwerasdf avatar Dec 08 '23 23:12 zxcvqwerasdf

https://github.com/lucidrains/BS-RoFormer

There is code, but no pre-trained model.

Is this what you are looking for https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/tag/v1.0.0? There is a pretrained mel_band_roformer model checkpoint, but it only achieves a SDR of 8.42. However, based on the SDR, I'm not sure if this will reproduce the same results as the sami-bytedance code from the SDX23 leaderboard C.

Ma5onic avatar Jan 10 '24 14:01 Ma5onic

Any updates on pre trained model?

AdamGoodApp avatar Mar 10 '24 04:03 AdamGoodApp

MVsep has a new good BS-Roformer model, it's, free to use on the website but the model is not publicly released (so can't be added to UVR).

jarredou avatar Mar 11 '24 01:03 jarredou

MVsep has a new good BS-Roformer model, it's, free to use on the website but the model is not publicly released (so can't be added to UVR).

SDR results seem pretty good, is it better than mdx23c?

thecatontheceiling avatar Mar 11 '24 01:03 thecatontheceiling

https://github.com/TRvlvr/model_repo/releases/download/all_public_uvr_models/model_bs_roformer_ep_368_sdr_12.9628.ckpt Requires use with this version of UVR: https://github.com/TRvlvr/model_repo/releases/download/uvr_update_patches/UVR_Patch_3_29_24_5_11_BETA_full_roformer.exe

XUANHLGG avatar Apr 04 '24 03:04 XUANHLGG

Let us know when theres a full 4-stem model of BSRoFormer! Colab if possible...

Dyslexicon avatar Apr 04 '24 04:04 Dyslexicon

https://github.com/TRvlvr/model_repo/releases/download/all_public_uvr_models/model_bs_roformer_ep_368_sdr_12.9628.ckpt Requires use with this version of UVR: https://github.com/TRvlvr/model_repo/releases/download/uvr_update_patches/UVR_Patch_3_29_24_5_11_BETA_full_roformer.exe

Does the Mac m1 have it?🙏

realzsan3 avatar Apr 16 '24 08:04 realzsan3

https://github.com/TRvlvr/model_repo/releases/download/all_public_uvr_models/model_bs_roformer_ep_368_sdr_12.9628.ckpt需要与此版本的 UVR 一起使用:https://github.com/TRvlvr/model_repo/releases/download/uvr_update_patches/UVR_Patch_3_29_24_5_11_BETA_full_roformer.exe

Mac m1 有吗? 🙏

Unfortunately, the UVR that supports BS Roformer is only available for Windows, and to use it on Mac, you may need to compile it yourself :(

XUANHLGG avatar Jun 20 '24 09:06 XUANHLGG