https://www.theverge.com/2019/11/5/20949338/vocal-isolation-ai-machine-learning-deezer-spleeter-automated-open-source-tensorflow

This open-source AI tool quickly isolates the vocals in any song
Perfect for making karaoke backing tracks and mashups

https://github.com/deezer/spleeter

------------


Gone: My Discogs collection for The Roots:
http://www.discogs.com/user/tomhayes-roots/collection

its actually surprisingly good

I have a few tracks I would love to sample but I don’t have the instrumentals.

Will give it a try.

****************
TBH the fact that you're even a mod here fits squarely within Jag's narrative of OK-sanctioned aggression, bullying, and toxicity. *shrug*

IG: amadu_me

"...Whateva, man..." (c) Redman

https://www.reddit.com/r/WeAreTheMusicMakers/comments/dubd2y/how_to_use_spleeter_on_windows_for_absolute/

I followed the above post and got it installed.

The results:
For some songs it works really well - for others not so much.

The vocal tracks never sound great - but some of the instrumentals do.

One Two Shit A Tribe Called Quest instrumental -
https://vocaroo.com/i/s0KArUq6jAFJ

One Two Shit A Tribe Called Quest vocals -
https://vocaroo.com/i/s080SNlcNR93

Hard Times by Baby Huey instrumental - https://vocaroo.com/i/s1XcPmRbnpyb

Metal by Gary Numan instrumental:
https://vocaroo.com/i/s0MPbStjiHfb

Nasty Habits - Oingo Boingo
https://vocaroo.com/i/s1LQnfVL7HHx



I think this could be useful for producers and casual Karaoke - but it's not magic.

------------


Gone: My Discogs collection for The Roots:
http://www.discogs.com/user/tomhayes-roots/collection

There's this thing called Google Colab, and it can handle installing and running these scripts directly. There are a few instances of this up right now (search google for 'spleeter colaboratory').

https://news.ycombinator.com/item?id=21431629 this one supports uploading from youtube as well as a direct mp3.

Simple instructions: Click the play button next to !pip install spleeter. When it's done, you may need to refresh the page. Click Option 1 or 2 on the side. If you click Option 1, hit the play button next to the 'from google.colab' line at the top. Then hit Choose Files and upload your song (named audio_sample.mp3). Choose between the 2-4-5 pretrained stems, then hit the play button at the top of the new cell. You're supposed to then be able to download the samples from that page, but it didn't work out for me. I had to click on the Files tab on the left, then download the audio files from the audio_output folders. Overall, very simple to use.

If I were into Karaoke, this shit would be crack. It's def going to be useful for blends and even production. The way it isolates vocals is nuts. Sure, it's not perfect, but you don't always need a lot. Just a simple clean vocal sample can make all the difference.

Here's a directory where I'm uploading stuff to:
http://www.tomsfiles.com/spleeter/

I did most as 2 tracks, but some have 4 or 5 splits.

Really enjoyed Gill Scott Heron's 1980 split (the fist part is weird sound effects - music starts around 30 in) - but the bass SLAPS and it's way better than how Tupac used it:
http://www.tomsfiles.com/spleeter/1980/Gil%20Scott%20Heron%201980%20insturmental%20(spleeter).wav

But if you're trying to find a break where none are open, or a small bass line, or just a loop - this could be really helpful.

------------


Gone: My Discogs collection for The Roots:
http://www.discogs.com/user/tomhayes-roots/collection

I tried a few tracks and I'm amazed by some of the vocal isolation. At first, it was a lil rough, but it sounds pretty good with some filtering. Give this to the right producer, and it might be a cheat code.

I can't wait to try grabbing some breaks, or isolating certain instruments. Remixes, or blends of songs that never had acapellas will also be great.