I’ve been demixing audio for a number of years, initially via Prosoniq Isolate and of course Spectralayers and sometimes Melodyne…whatever works best for a given project.
The Spleeter approach has been nice in Spectralayers…but geez…I was just listening to a bunch of Demucs-based technology and so far, to my ears, it seems massively superior to spleeter results
Seems to me that Spectralayers would benefit greatly.
I had just extracted mono drums in SL Pro 7 (from one of my 1966 recordings where I no longer have the multitrack tape) and was trying to figure out the best way to now separate the bass drum, hh, snare, sidestick, ride cymbals, crash onto separate mono audio tracks for a rebuild in Cubase 11 (an unmix drumkit preset that would be fantastic to have in SL !!). At any rate, a google search to get some ideas (rather than manually extracting the way I usually do) took me almost immediately to demucs topics…where drum extractions are soooo much more clean/less artifacty than spleeter. By several magnitudes imo.
It doesn’t look like anyone’s developed a workable plug-in GUI for Demucs. A lot of people seem to be interested in Demucs, but real-world implementation appears to be a sticking point.
Based on what I’ve seen, no matter how good the processor tech is in theory, it still has to be simple enough to drop into an audio track and work with just a few faders. Otherwise, it defeats the purpose of getting things done as quickly as possible. Spleeter may not be perfect, but devs have at least made a marketable product out of it.
I was initially really impressed by the unmix stems function, unmix drums in particular, and bought Pro for this as I live record a lot and thought I’d use SLP7 for debleeding my drum tracks. However, I have given up on this since the unmix function leaves a lot of artefacts and actually seems to have a lot of trouble even separating out the drums from the overhead track, making them sound phasey and leaving lots of the bleed from other instruments in.
Also, SLP7 doesn’t do a good job at debleeding even close miked drums, like a bass drum.
I would LOVE it if SLP could be adapted to this kind of functionality in the future. Right now, I only use the unmix for vocals, which it is kind of good at - even though I think it still leaves the unmixed stem somewhat phasey.
Also, I really really think SLP needs a much better manual. The current one is too superficial to be useful.