Yeah by now I was hoping that we’d be seeing more “upscaling” of audio processes like we do with images.
Of course this could probably add as many artefacts as remove but I’d be interested what it could do with a separated vocal stem, re-adding lost harmonics and sibilance etc.
Is this a new one ? 14 stem types sounds very different to the recent posts I read here.
Comparing splits I’ve done in Koala vs Logic Pro for iPad, generally the logic ones come out a bit cleaner. While I suspect it’s the same tech under the hood, I think Apple have tuned the algorithm a bit.
Koala is using Spleeter while the latest splitter algo is demuc 4 ht. Have feature requested for an upgrade.
Managed to install StemRoller. Big difference in-between it and Demucs-GUI is that StemRoller is 2,2GB large app while Demucs GUI is only 500Mb. Demucs runs locally with MPS (Mac GPU support), not sure about StemRoller. In Demucs you can choose the specific stem you want to extract and not spend time on all. Demucs GUI doesn’t claim to be AI when in reality it is just an algorithm and it doesn’t run some shady connections.
Logic is dope… new update made alot of my third party programs obsolete… Im also not that much of a stickler for “the best…” some artifacting lends itself to texture
Demucs uses a ML library, so that’s not an inaccurate claim. The training model is local, it’s “AI” similar to Topaz video cleanup etc. “ML” and “AI” were used pretty confusingly interchangeably until generative slop companies started polluting the public understanding of what ML is.
Hrm, the latest Demucs GUI was 5GB for me fully un-7z’d on a Windows box with CUDA.
I’ll probably stick with that instead of Stemroller still, thanks!
I know the app had a lot of bells and whistles to let people download from Youtube but like, I don’t need any extra features baked in so fine with getting rid of the bloat.
Apologies if this has been mentioned already. I’m using Splitter from Bandlab.com. Works for me. Never invested any time comparing it to other solutions. Stems are getting highly mangled anyways, hence no need for perfect quality.
As a Serato user, I was thinking the other day of buying Serato sample to use in Ableton Live. But after watching a workflow video, it made me realise that I could just use my Serato DJ software to grab out the stem split files.
I’m going to give it a try tomorrow as I’ve always been pretty happy with the stem splits from Serato.
You can here the difference in-between Spleeter and Demucs 4 ht here. Better extraction of, in this case, vocals and retaining the full range without beat leakage:
Original
Spleeter (Koala app)
Demucs 4 ht
This sounds great. I’ve just bought a Mac mini m4 and would love to use it to split tracks. Is there an “easy” demucs 4 (ht?) installer with gui you can link me to? I’m a Mac noob and keen not to go into any compile type of business…thanks!