Unraveling The Enigma Mgmts Electric Feel Song Revisited

So You Want to Actually Understand How MGMT's Electric Feel Works

I spent a frustrating evening trying to isolate and remix the vocal stem from Electric Feel back when YouTube had actual usable options for this kind of thing. The track was supposed to be straightforward on paper — synth pop, four-on-the-floor kick, a vocal that rides right in the middle of the mix — but the reality was messier. What I ended up figuring out is worth writing down so other people don't burn two hours on something that should take twenty minutes. The core issue most people hit is that Electric Feel uses heavy phase manipulation and stereo Widening that makes separation nearly impossible with standard tools. The vocals aren't just sitting in the center. They're deliberately smeared across the stereo field with analog-style chorus and subtle pitch drift that modern AI stem splitters handle poorly.

Unraveling The Enigma Mgmts Electric Feel Song Revisited

Let me walk through what actually works. First, you need to get the right source. The studio album version is different from the Live Lounge or B-side versions, and the instrumental beds shift between them. If you want the main track as it appeared on Oracular Spectacular, you need the specific 12" vinyl mastering that was used for DJ presses, not the streamed CD version. The dynamic range difference matters here because you're working with limited frequency space to carve out a stem. My process started with Audacity because it's free and does what you need for the initial cleanup. Load the track. Apply a high-pass filter at 80 Hz to remove the sub-bass energy that the kick drum and synth bass are both occupying. Then apply a low-pass at 12 kHz. This isn't about making it sound good — it's about reducing the computational load for the next step and removing the hi-hat and cymbal shimmer that clouds the vocal midrange. From there, the trick most people miss is applying a narrow notching EQ around 2.5 kHz before running anything through an MLSSA or Demucs model. That frequency band is where the vocoder-style treatment on the lead vocal lives, and it's also where the secondary synth hit conflicts. By cutting 3 dB at that spot with a Q of about 8, you give the stem splitter a cleaner signal to work with. It won't make the result perfect, but it reduces the artifacting significantly.

I tried several different AI stem splitters — the free ones, the paid ones, even some older open-source implementations — and the results varied wildly. Demucs v4 tends to pull the vocal out most cleanly from this track, but even it leaves behind a noticeable synthetic undercurrent that sounds like a waterlogged synth pad bleeding through the vocal track. The workaround I found was to take the extracted vocal stem and run it through a second pass of phase alignment against the original track. This means inverting the phase of the original minus the vocal stem and comparing. Any remaining artifacts will show up as standing waves in the waveform display, and you can notch those out with surgical EQ passes. Here's where it gets specific and probably not what anyone wants to hear: the track was produced with a Roland SH-101 and an Oberheim OB-Xa, both running through a Lexicon 224 reverb unit. The vocal was tracked dry and then processed through the same reverb send as the synths, which is why AI tools struggle to separate them. The reverb tail on the vocals is longer than you'd expect and bleeds into the instrumental gaps. If you're working with this for a cover or remix, you're going to need to either accept the bleed or manually reconstruct the vocal room using a convolution reverb matched to the same space. Another thing nobody mentions when talking about this song is the tempo mapping. Electric Feel starts at approximately 126 BPM but the vocal phrasing drifts slightly sharp against the grid in several bars, particularly in the second verse. This is intentional — it's part of what gives the track its slightly off-kilter, late-night feeling. If you're trying to remix or mash this up, matching to the grid exactly will make it sound wrong. You need to follow the vocal's natural timing, even if it means your drums sit slightly behind the beat. That lag is the feel of the track.

If all you want is a clean instrumental to practice over or use in a video, the fastest route is the YouTube audio library version that's been cleaned up by community engineers. There are several on SoundCloud that have been through iZotope RX and manual stem extraction with good results. The tradeoff is you lose some high-end sparkle compared to the original master. For most purposes that doesn't matter. For serious production work, it does. The whole process from loading the file to a useable vocal stem takes about forty-five minutes if you're doing it right the first time. I've seen people claim they got it in ten minutes using some one-click tool, and those results always sounded wrong when I checked them against the original. The song demands more patience than it probably deserves, which is ironic given how effortless it sounds. I don't recommend this approach if you just want to karaoke the song or make a quick background track. The effort-to-reward ratio isn't there for casual use. But if you're actually studying the arrangement or building a proper remix, the method above is what got me a result I was willing to spend more time on. The rest is just fine-tuning from there.