Maybe it’s a result of my previous searches, but it seems like I can’t open Facebook these days without being bombarded with stem separation solutions — and FADR is probably the most publicised.
So, after a barrage of offers, I finally succumbed and grabbed the paid version, as the free one didn’t really give an indication of what the solution entails.
And for those who haven’t seen it previously, digitalDrummer has been reviewing stem separators since they first emerged, partly in search of the perfect drum-free play-along creator. Just search our archive and you’ll see our previous investigations.
How It Works
FADR is an online suite covering a range of musical tools. The basic free programme includes a stem separator (vocals, drums, melodies and bass) and allows you to download the basic stems as MP3s.
So, nothing amazing there — but perhaps enough to encourage you to sign up for the paid FADR+ subscription, currently US$10 per month or $100 per year. For that, you get extended stem separation capability (16 stems), high-quality .wav downloads, MIDI detection, a Remix Maker and access to an API so you can work on your computer rather than doing everything online.
In Action
If you’re using the online version, obviously, you need an internet connection — and the faster it is, the quicker you see results.
Maybe I’m just becoming an old hand at these, but the solution is easy to use; the steps are clear and straightforward, and you can either drag and drop songs or search and load them. Once you select a song, it automatically begins to stem, and you soon have the BPM, key and four stems. For a complex song like Toto’s Africa, it took just over a minute to grab another 10 stems including separated kick, snare and hats.
The stems are presented in a DAW-like arrangement, each with a Solo/Mute selector and volume and pan sliders.
When you’re happy with the balance and, for example, the drum removal, you can export the stems individually, export the MIDI or hit “Remix”. That function isn’t really explained, but it seems to fill out some of the gaps left by the separation process. In case you’re new to this, the separation process is a bit like unscrambling an omelette — it’s not entirely possible to separate all the components without some contamination: a few kick drum frequencies are captured as bass guitar, some hi-hat hits are dropped, etc. So when you put it all back together, there are fragments missing and other bits exaggerated. Hopefully, none of that damage detracts too much from the overall listening experience.
For our review, we used our stock tracks — a selection of songs, each presenting a particular challenge.
- Toto’s Africa: A busy song with lots of instruments playing simultaneously — drums, percussion, keys, guitars and multi-part vocals. This is the first tool I’ve used that extracted the hi-hat on a single stem, and it did an impressive job, grabbing all the hat hits. It also included the cowbell hits — and there were plenty of those — in the hi-hat stem. Similarly, the bass stem included some lower percussion tones, and the snare stem grabbed some mid-range Latin percussion. While none of the tracks was totally pure, muting them all produced a very credible drumless track. Grade: A
- The Shadows’ Apache: We always include this one because the snare is so closely linked to the rhythm guitar, with the two seemingly merging for much of the track. Interestingly, as a drum stem separator, the tool worked better in basic mode than advanced. I got a good, crisp snare in the mixed drum track but a very thin version on the snare stem of the three-part split. The signature toms were scooped up by the bass filter — in case you wonder where they go. And the hi-hat stem captured some snare and all of the ride. So, as a drum part learning tool this didn’t quite cut it, but for a drumless track it warrants a B.
- Elle King’s Ex’s & Oh’s: This is on our test list because a lot of floor tom is played on top of the bass guitar. The experience here was pretty much the exact opposite of Apache. The solo drum track (before it was split in three) was almost perfect. When it was split again, the app clearly struggled — the predominance of tom hits meant most were read as bass drum strikes. And because of the overlap between toms and bass guitar, the drumless track left the bass somewhat depleted. So, for learning the drum part, it’s an A, but for the more probable use as a drumless track, FADR gets a C.
- The Beatles’ When I’m Sixty-Four: Using an MP3 version of the remastered track, I got a pristine drums stem with the brushes and kick neatly isolated. The bells were captured in the “instrumental” stem, along with the wind instruments. The basic separation was impressive — both for the isolated drum reference track and for the drumless output. The advanced separation didn’t enhance the drum tracks at all, though it did produce very accurate backing stems. Grade: A+
- Santana’s Smooth: Finally, we tried this busy, percussion-heavy track with many instruments layered over each other. Using the basic separator, the resulting drum/percussion stem was clear and clean, and when muted, the drumless track remained relatively intact. By this point, I was noticing a trend: for drumless tracks, the free version was sufficient. The additional separation from FADR+ didn’t help for drum part learning and actually removed less percussion from the drumless track when muted. Using the free version, FADR gets an A as a drum stem separator.
Overall
You can thank me later for saving you $100 — if you want to produce drumless tracks, don’t bother with a subscription.
And remember that with stem separation, the usual caveat applies: some mixes are easier to unscramble than others. For most of our test tracks, the free FADR produced usable drumless tracks. Were they better than other free offerings? Not really. I suspect almost everyone is using the same stem engine — a derivative of the original Spleeter script — and I’m not sure the producers are throwing in AI references for any reason beyond capitalising on the technology’s current popularity.
What about Drum GPT? FADR flags this as “an AI drum generator that turns text prompts into 16-note drum kits”. The 16-note claim is a bit overstated — several kits I generated allocated the same sounds to toms 1 and 2. And on the accuracy front, the kit conjured up in response to my request for a “1967 Ringo Starr tea towel Ludwig kit” was not too different from the one offered for “Phil Collins’ In The Air Tonight gated Gretsch kit” — both being very far from the mark. VST developers have nothing to worry about.
FADR may deliver plenty of value to DJs and music remixers, but for drummers there’s not a lot there — and certainly nothing worth paying for.