Genres like metal or black metal that includes electric guitar are detected as melody stem? For my device instrument stem is renamed just to melody and I wonder about guitars? How Some of the high notes and low notes works for you?
Just try it. You could also ask if the cello, trombone and euphonium of an orchestral piece are taken in the bass or melody part, but you can imagine that AI will make mistakes…
Yesterday I heard a STEMmed Offspring track which had good separation. I also heard some dance tracks where melody parts where mistaken as vocal.
Its a shame that there cant be a detailed track by track algorithm database of some sort, where vendors can tap into data done professionally on a track by track basis, which is then compressed into some kind of tiny binary code file that can be passed to the device to perform the rendering.
A massive pipe dream at this point, and one that would require resource to manage, but it would be awesome if doable.
1 Like
Licensing issues with intellectual property. You run into this everyday with music, but the reality is that people are dead scared of other people stealing their hard work.
I gave the orchestral example above for giggles, but now you start on IP: well, last year I started to play double bass in a concert band. The benefit here is that everything is nicely written out in contrast to rock music (been playing electric guitar for 10 years now), but these orchestral scores come pretty expensive. For our next concert, the band has apparently bought worth of around 1000 euros of scores.
Now, for some pieces, one of our members, who is a music teacher, has written out arrangements himself. The problem here is that you wont find any recorded versions of this, and as a beginning double bass player who isn’t very good at sight reading, I sure would have found an audio rendering super-handy for practicing at home. And I am sure he can just click on “render as audio file” in his scoring software. But the answer is nuts. As much that I am frustrated about this I do understand that he can sell his arrangements to other bands after our concert for a pretty penny. If a recording leaked, other people could start transcribing this and sell it as their own, before he published his own score.
Or in some parts of an orchestral piece you don’t really grasp what the other instruments are playing (on the downbeat or after the downbeat, counting to 4, or to 3 or even to 7), which makes it difficult to follow along and know when its your turn to start playing: your score only shows bass. The directors score shows all of this, but don’t dream of lending this copy: should this copy start to circulate the internet our concert band is in big trouble, because these orchestral scores are actually stamped with the name of the buyer on it to prevent theft 
. And understandably, if you can ask 100 euro for 1 orchestral score, you don’t want local hobby orchestras to start passing your IP around to each other!
Just to give an example why any information about how a track is put together is a well guarded secret. Offcourse, anyone could transcribe it, put then you havent got the exact synth patch, you cant sample a transcription, and so on…
This is one reason why the original Traktor stems never took off. It was highly unlikely that record companies and/or artists would be up for distributing the component parts of their copyrighted material to all and sundry.
The other problem of course is that it would take forever (and be costly) to go back to all their original multitrack recordings, take them back in the studio, pay an engineer to sit and mix everything down to just four tracks that were still perfectly blended to sound correct when played back together, then package them up and sell them to a restricted market.
I know some of the other more pro stem separation apps can target individual instruments - but it’ll cost you more than 9.99 and also we’ve got no way of saving to Denon’s format.
- RipX DAW Pro = ÂŁ198
- SpectraLayers Pro = ÂŁ248
1 Like
Yeah, I think we can conclude AI is the only way stems were ever going to happen. Allthough this also shows AI is a big threat for the music industry. I already had my 2 cents on what AI (music) creation would mean for the value of artistic work, but it works the other way around too. You already have AI transcription tools, stem separation, recreate a synth patches, guitar amp patches, and so on….
Oh, and yesterday there was a punter who wanted me to play an EDM track which he created using AI. And honestly, apart from the mediocre mastering the result was on par with bigger producers! If he would have told me he made it himself I would have believed him!
How are you, as a creator, going to defend your work from theft when these tools get refined? How are you going to be appreciated as an artist when the computer does the job just as well? How are you as a DJ going to get paid 250 quid for playing all night long if Spotify has an AI DJ in the works? This is a rabithole…
Now, the evil upside is this: Western society has become rather lazy, where all the hard work is delegated to eastern countries, and instead we invent the most stupid administrative jobs for ourselves, just to fight boredom. In the philosophy of “creating” and “displacing” added value, this is an economic nightmare. Thanks to AI however, we can have ridiculously complex rules AND do hand labour at the same time. Solves obesitas too! Actually every intellectual work can eventually be replaced by computers… Expensive politicians are finally a thing of the past! The only downside is that society will be ruled by computers. Well, computers are probably more dependable than human politicians too (until these computers watch Terminator…)
I think you mean a big threat LOL 
Yes, even spellcheck will be better using AI! 
Yeah I have also Spectra Layers 10 as advanced spectral editor updated to 11 version and these reverse engineering tasks are easy when the audio engineers want to do audio restoration and fixing some unwanted sounds/frequencies from mix. In the whole Un-mixing process it’s very difficult for some complex tracks and painful without someone help and it can be tricky for ear catching. I’m not engineer but these softwares for that should be used with other analog equipments especially for unmixing than restoration and fixing. In the DJ workflow using this staff not working because there’s no any batch mp3 or wav extraction for multiple files at once, more stems of single track means more available decks and channels to play more individuals parts from tracks.
Hopefully in the near future Google brand new Willow Chip based on Quantum computing can handle this for music hardware or OS devices only when they release this product for everyone.