Post-challenge discussion

@XavierJ thanks for the discussion. I can just respond to one thing, as I was directly involved with it

I received an email from the organizers asking whether we could provide an implementation of BSRNN to serve as a baseline for the challenge. We did submit one system and made the entry publicly available (Tomasyu / sdx-2023-music-demixing-track-starter-kit · GitLab, submission #209291), but it seems that the organizers haven’t marked it as a baseline like several other baseline models until now. We did not get followups about whether the info about this system has been shared with any participants since we submitted it.

I was in contact with Jianwei Yu from your team via mail in Februrary. I think after the last issues have been resolved with the submission, we already entered phase II and you didn’t resubmit (which means the submission dissappeared). I then forgot to ping the AIcrowd again then. So I’m sorry for that.

I will make that we rerun the submission again so that it appears on the final leaderboard and will be marked as a baseline with appropriate links. Is that ok?

View post 0

Demixing Challenge Workshop

Hi, I just want to remind everyone again that we are calling for participants for the music demixing workshop that will take place online on November 12th. Everyone who submitted to the ai crowd challenge is very well invited to submit a poster, talk or discussion. It doesn’t need to be state-of-the-art!

The MDX21 workshop (https://mdx-workshop.github.io) is a satellite event at ISMIR 2021 and is the official academic follow-up for the 2021 music demixing challenge organized by Sony on aicrowd, to which you participated. It will feature invited talks as well as presentations and posters, on the topic of music processing.

__The abstract deadline is next Thursday October, 28th __

The submission system is open at the following link:
https://github.com/mdx-workshop/mdx-submissions21

We accept three kinds of submissions:

1. Posters. Everyone may submit a poster, notably you that participated in the Music Demixing Challenge. After a minimal prescreening, you would have the opportunity to present your work in the online virtual space for the conference. Pick the POSTERS category, and submit a title + short abstract.

2. Long presentations (20min+questions), during which you can present a recent research or some topic you think could be of interest to the community. Pick the LONG TALK category, and submit a title + extended abstract for your talk.

3. Discussions (30min), during which you propose to initiate and moderate a group discussion about a particular topic after a 5min introduction. The objective is to stimulate new ideas and collaborations on music separations. Pick the DISCUSSIONS category, and submit a title + extended abstract that describes discussion topics.

Relevant topics for submissions include:

• music demixing / source separation
• representation learning for musical instruments and mixtures
• generative models for music
• applications of music demixing and filtering
• self-supervised music processing
• deep neural nets for very long waveforms
• machine learning with applications on music stems and mixture signals

View post 3

Due Diligence

yes, see my answer above. you can do this

View post 1

Due Diligence

@agent I am not a lawyer, so I can’t give you a legal advice. My understanding is that model code != weights. So, if you loaded the pre-trained umxl weights and used the results for a model fusion, this would be considered as modification of the weights itself. So, your code can be MIT as long as you did not export/redistribute the umlx weights as part of a joint model.

Ideally, your code would load the pretrained model from torch.hub directly.

as I understand, in the following Example it would not be allowed to redistribute fusioned_model under MIT license.

# load umxl weights CC4.0 SA licensed
model1 = torch.hub.load('sigsep/open-unmix-pytorch', 'umxl')
# load other MIT licensed model
model2 = torch.hub.load('user/other_model', 'other_model')
fusioned_model = torch.nn.Sequential([model1, model2])
torch.save(fusioned_model, ...)

however, you can of course create any MIT licensed code that uses model1 as part as your algorithm.

Sorry for the confusion, does this make sense?

View post 1

Due Diligence

@agent if the model is modified or adapted (e.g. you created a new model based on umxl), the answer, unfortunately, is: no (see https://creativecommons.org/share-your-work/licensing-considerations/compatible-licenses/)

View post 1

Due Diligence

@agent again, if you did not fine-tune the model using additional training data and managed to improve the scores due to blending I don’t see any reason why you would need to add info on training of umxl for the sake of reproducibility.

Still, your description is accurate, the umxl model was trained using the normal training scripts. umxl is double the hidden_size (1024) compared to umxhq. Other than that, we used the same number of gpus (one for each target), the training time was about 2 weeks on 2080 gpu.

View post 1

Due Diligence

@agent oh and: congrats!

View post 1

Due Diligence

@agent we can’t provide you with a tracklist. Since umxl was published under an open-source license, I don’t see a problem with reproducibility as long as you didn’t fine tuned the model

View post 1

Alternative download mirror for MUSDB18-HQ

We appreciate your feedback! If these problems would have happened in the beginning of the competition we would have tried to find an alternative mirror - but I guess everyone managed to get a copy now

View post 1

Alternative download mirror for MUSDB18-HQ

I’m sorry, that you are having such a bad experience with zenodo. I just tested the download and I get around 2 Mb/s, so if there was an issue, it has been solved.

sigsep/musdb18 is a community effort. We have to rely on free services, if you know a better suited alternative than zenodo, please let me know. We tried also https://academictorrents.com/ with little success…

View post 0

The leaf - wicked drums file from musdb18 has the electric guitar playing on it

@kimberley_jensen I started here: https://github.com/sigsep/website/pull/27

feel free to add more things in the PR

View post 2

The leaf - wicked drums file from musdb18 has the electric guitar playing on it

@StefanUhlich @kimberley_jensen okay, i will start an errata list here: https://github.com/sigsep/website/blob/master/content/datasets/musdb.md

and will invite you there.

View post 2

Question about MUSDB18

Hi @agent thats a good question: The reasons for this were:

• the remaining MedleyDB tracks all contains bleeding. This may or may not be a problem for training we didn’t wanted to make it more risky
• there are a lot of more stems from the native instruments stems pack but most of them did not contain any vocals.

View post 3