__Repository of old Colab notebooks__
UVR 5 (Colab by HV): https://colab.research.google.com/github/NaJeongMo/Colaboratory-Notebook-for-Ultimate-Vocal-Remover/blob/main/Vocal%20Remover%205_arch.ipynb
(On Mobile Chrome use PC mode)
Alternative UVR 5 notebook up to date (not HV’s):
MDX (Colab by CyberWaifu, 4 stem, cannot be used in Mobile Chrome even using PC mode - there's no GDrive mounting and track downloading is always 0%. Model A cleaner but with more bleeding; Audioshake is based on it, but with different model based on larger dataset iirc, UVR team consider training it on their own bigger dataset to get better results - it’s based on phase unlike UVR, but tsumeruso works on adding phase, so then it might get rewritten to UVR)
https://colab.research.google.com/drive/1R32s9M50tn_TRUGIkfnjNPYdbUvQOcfh?usp=sharing
(wait patiently, it doesn’t show the progress)
UVR 5 (old version by HV with any 2 files ensemble feature, put tracks in separated folder. As for x/z - similar results, but not the same. Put as first the one you want the result more similar to)
https://colab.research.google.com/drive/1eK4h-13SmbjwYPecW2-PdMoEbJcpqzDt?usp=sharing
https://colab.research.google.com/drive/1C6i_6pBRjdbyueVw27FuRpXmEe442n4k?usp=sharing#scrollTo=CT8TuXWLBrXF (+12 ens, no batch ens, deleted)
2021-ISMIR-MSS-Challenge-CWS-PResUNet (byteMSS) (if you run out of memory, split up the input file)
https://colab.research.google.com/drive/17m08bvihZAov_F_6Rg3luNj030t6mtyk?usp=sharing
Woosung Choi's ISMIR 2020 (Colab by CyberWaifu)
https://colab.research.google.com/drive/1jlwVgC9sRCGnZAKZTpqKgeSnzP3sIj8U
Vocal Remover 4:
https://colab.research.google.com/drive/1z0YBPfSexb4E7mhNz9LJP4Kfz3AvHf32
To fix librosa error, try adding the
!pip install librosa==0.8.0
or 0.9.? works as well
line about librosa, and if still the same, about pysound as well:
https://discord.com/channels/708579735583588363/767947630403387393/1089518963253317652
(UVR4 + models description:
https://github.com/Anjok07/ultimatevocalremovergui/tree/v4.0.1
Search for:
"Models included" at the bottom".)
UVR 2.20 (it achieved some good results for old 70’s pop music for me where cymbals got muffled on current models, but prepare for more bleeding in some places vs VR4 and newer)
https://colab.research.google.com/drive/1gGtjAo3jK3nmHcMYTz0p8Qs8rZu8Lhb6?usp=sharing
Spleeter (11/16kHz, 2, 4, 5 stems, currently doesn’t work): https://colab.research.google.com/drive/1d-NKFQVRGCV5tvbd0GOy9spMMel6mrth?usp=sharing
According to my experience, if you don’t need piano stem, 4 stem model makes better job than 5 stem (and even vs 2 stem, and it is also reflected in SDR results). Use 11kHz models only if your input files are sampled at 22kHz (it will provide better result in this and only in this case).
If you can, use Izotope RX-8 for 22kHz 4 stem, as it provides better separation quality with aggressiveness option. It’s Spleeter, but with better model (full band).
Demucs 3.0
https://colab.research.google.com/drive/1yyEe0m8t5b3i9FQkCl_iy6c9maF2brGx?usp=sharing
To install it locally (by britneyjbitch):
I cracked the Da Vinci code on how to install Demucs V3 sweat_smile For anybody who struggled (on Windows) - I got you!
1. DL a zip folder of Demucs 3 from Github (link: https://github.com/facebookresearch/demucs) and extract it in a desired folder
2. Inside the extracted folder run cmd
3. If you want to simply separate tracks, run the following command:
python.exe -m pip install --requirement requirements_minimal.txt
4. If you want to be able to train models too, run the following command:
python.exe -m pip install --requirement requirements.txt
5. If a read error for incompatible versions of any of the modules appears (e.g. torch) run the following command:
pip install desired_module==version_of_desired_module
e.g. pip install torch==1.9.0
6. Repeat step 5 for any incompatibilities that might occur
7. Separating tracks:
python.exe -m demucs -n "desired_model_to_run_separation" "path_to_track"
8. If you want help finding all additional options (for example overlap or shifts), run:
python.exe -m demucs --help
At least that worked for me, feel free to let me know if this worked for others as well
exclamation Oh, and I forgot - between step 6 or 7, don't pay attention to a potential red error ''torchvision 0.9.1+cu111 has requirement torch==1.8.1, but you'll have torch 1.9.0 which is incompatible.''
Do NOT change back to torch 1.8.0 cuz you won't be able to run demucs
warning! If ''torchvision 0.9.1+cu111 has requirement torch==1.8.1, but you'll have torch 1.9.0 which is incompatible.'' is the only red error you're getting after executing the commands from step 3,4 and/or 5, you're good to go with separation!
Demucs (22khz, 4 stem):
https://colab.research.google.com/drive/1gRGRDhx9yA1KtafKhOaXZUpUoh2MuF8?usp=sharing
https://colab.research.google.com/github/facebookresearch/demucs/blob/master/Demucs.ipynb
https://colab.research.google.com/drive/1gRGRDhx9yA1KtafKhOaXZUpUoh2MuF_8?usp=sharing
Other one(s):
LaSAFT:
https://colab.research.google.com/drive/1XIngzXDi2mF_y6WwDrLLx4XZtI8_1FAz?usp=sharing
(original, cannot define model ATM)
If you cannot load the file, upload it manually to your Colab, or just wait patiently. Refresh Github page with CTRL+R if you can’t see the code preview.
Check out also this laSAFT download with message which says about superiority of 2020 model (said in march 2021).
Clone voice:
Matchering:
For more Colabs search for colab.research.google.com on our Discord server