Sound Design Stack Exchange is a question and answer site for sound engineers, producers, editors, and enthusiasts. It only takes a minute to sign up.
Sign up to join this community
I have a track I wish to hear without the vocals/bass/drums.
How do I do that?
This is an attempt to make a new canonical QA based on current technology & ignores 'old-style' manual phase reverse tricks etc
It can be post-linked to other questions as we find them.
I'm aware it might be considered a bit of a 'list question' but I feel that might be useful in this instance.
I've marked my first answer as 'accepted' just to give the question a green flag when people search.
These days, what used to be referred to as "the holy grail of the misinformed" is getting closer to actuality.
There are now applications designed to do exactly this task - isolate parts of a complete mix & allow you to change their relative levels; including total removal or isolation of sections such as vocals, drums, bass or 'instruments'.
Results are not perfect every time, but are getting reasonably close these days.
The most comprehensive of these so far is iZotope RX [curently at v8, though I only have v7 to show images from] Various versions from 99 to 1000 USD.
iZotope has many audio cleanup functions, but here is the Music Rebalance plugin, which can isolate 4 types of music data - Voice, Bass, Percussion, Other.
After that, a new kid on the block [Mac only at the moment] is Neural Mix Pro, free download, but 50 USD to enable the save function.
It's less comprehensive than RX, but it is real-time & needs no pre-processing.
Results can be a bit phasey but it doesn't do too bad a job.
It opens by default showing your iTunes music library, but you can drag & drop files to it too.
It can do other tasks, somewhat DJ-oriented, such as time-stretch & pitch-shift but they're not under consideration for this QA.
As new products become available, please add to this starter list.
Deezer open-sourced their Spleeter stem separation engine in November 2019: Announcement here and the application is available on Github here.
There's also Acon Digital's Acoustica here which since version 7.2 has offered stem separation tools (based on the Spleeter algorithm).
Steinberg SpectraLayers from version 7 offers track to stems extraction (vocals only for Elements edition) and components extraction (tone/noise/transients) - underlying algorithm unknown.
I did a weekend hack to build a web interface for the Spleeter. I think it works fine for now to separate vocals and instruments. You can try it out here:
I did the bass backing track generator as well. It's there on your profile page when you're logged in. Basically, it uses the same tool under the hood. I would like to get feedback on it. For now, I just use the tool for my own purpose.
The most basic and "traditional" technique is mid-sides processing. The idea is that instead of the stereo signal being divided into left and right signals, it's divided into "middle" and "sides" signals. The easy way to think of it is
Separating this (and rejoining them back to left-right stereo) is just maths, with no possible user intervention. It is 100% reliable, repeatable and (most importantly) reversible.
Once you've got that, you'll find that lead vocals, bass, kick and snare all occupy the mid channel. At that point you can use EQ to cut the frequencies for the instrument you want to remove.
Current technology may give better targetted EQ, but ultimately this is almost certainly what existing plugins are doing under the surface.
This will remove all audio information that was mixed in the center (equally loud in both left & right channel). Since at least vocals and bass are usually mixed center, they will be removed although some reverb tails might still be audible. This is how the magic 'karaoke-button' used to work in some home stereo systems. This is easy and simple to do in any modern DAW software but has some nasty downsides: the result is mono and you can't choose what to remove - everything in the center will go.
This is partly what Graham is talking about but more in layman's terms.
