FM Soundwise

It's not uncommon to lose mixdowns, premasters, or even entire projects, leaving us with only the final master files - yet we still want to work on those songs, most often to remix and remaster them. After successfully reconstructing the audio of several major projects over the past year, I've decided to finally give this service a name and offer it publicly. As the word 'reconstruct' suggests, it involves reverse-engineering a song using modern audio stem-splitting technology, allowing for a seamless remix and remaster of a mastered stereo file.

Let me break it down in detail. Stem-splitting technology has been around in some form for about 10 years, but until just a few months ago, it was never convincing enough for professional audio work. Even today, some of the most well-known tools, like iZotope RX, still don’t quite do it justice. After waiting ages for the stem-splitting process to finish, you often end up with stems that have a decent amount of bleed between them - so you're not really left with fully usable material. What changed the game for me was the latest update of Logic X Pro where they introduced stem splitter.

On the very same day the update went live, I received one of the biggest requests of my mixing and mastering career - from none other than Edo Maajka, arguably the No. 1 rapper in the Balkans, and his label Menart. They asked me to remaster his legendary 20-year-old album No Sikiriki both for vinyl press and digital streaming. It was a pleasant kind of pressure and a huge motivation to give my best in bringing the sound of this old album into the present. And while it already sounds pretty darn good for a 20-year-old production, there was still plenty of room for improvement. As I started listening and analyzing all those hits, I had a strong urge to dive into every musical element, to rebalance the drums, basslines, instruments, and especially vocals. But the original projects and mixdowns were lost in time and all I had were full-blown masters, some pushing as loud as -4 LUFS.

I started working on the intro track Sevdah O Rodama, which consists only of a guitar and Edo’s vocal. At that point, I hadn’t even considered using Logic’s stem split option, assuming it wouldn’t be good enough for me to actually use. Since the track wasn’t crowded with elements, I was able to control quite a few harsh resonances in both the vocals and the guitar. I also wanted to bring up the fundamental lows of the very thin-sounding guitar - boosting them by around 8 dB. That worked well, but along with the guitar’s low end, I also brought up some very audible vocal plosives that hadn’t been noticeable before. Not doing low cut filtering on vocals 20 years ago was not a big surprise to me. That’s when I got the idea to try using the stem split - just to create a separate vocal channel, isolate the plosive frequency range, and then flip the phase polarity to cancel them out in the main file. It worked perfectly, and the intro track was done.

The vocal stem sounded surprisingly good on its own, which made me even more curious to explore it further on the next tracks from the album. I jumped into the next track To Sam Ja, and right away I noticed that Edo’s vocals were very midrange-heavy and resonant, making him sound slightly detached and sitting above the hip hop beat, which itself felt a bit muddy. On top of that, there were some arrangement issues - most notably, the main bassline was randomly disappearing by about 70% in volume due to some phase cancellation on every few hits of the bass line throughout the entire song. In comparison to Sevdah O Rodama, this was a full song with drums, melodic samples, bass, and vocals. And aside from the issues I already mentioned, I had a strong desire to dive even deeper - so I started experimenting with the stem-split drums, bass, other instruments, and vocals, and was genuinely amazed that there was virtually no audio bleed. I was still a bit skeptical, though, about the slightly 'watery' sound artifacts present in each individual stem. However, when the stems were summed back together in playback, those artifacts seemed to reconnect within the audio and became inaudible again.

That got me curious: how much could I actually EQ, compress, widen, or even saturate each individual stem before the artifacts became audible once everything was summed back together?

So I ran a longer test on To Sam Ja, making some drastic changes, and I was blown away by how far I could push it. I was basically able to remix the entire song without introducing any noticeable artifacts. On top of that, working with separated stems gave me a whole new level of headroom, allowing me to push certain songs even louder by applying individual stem clipping, limiting, and other dynamic processing.

Coming back to the phasey bass issue in To Sam Ja, since I had the bassline isolated in its own stem, I was able to simply cut out the clean bass notes and replace the problematic ones throughout the song. This is a great example of how stem splitting, beyond just mixing and mastering, can also be a powerful tool for fixing arrangement issues within a track.

After completing this first complex remix and remaster using split stems, I had full confidence to continue working on the album this way, and the rest is history.

The next big project was remixing and remastering of another legendary 20-year-old album, which is yet to be released by Menart and talked about a lot very soon. Again, I used Logic’s stem split feature to do a stem remix and remaster. Just to mention a few more of the techniques this approach made possible: deeper transient detection and manipulation on the drum stem with plugins like Spiff or Split EQ, more advanced mid-side EQ on all stems, seamless sidechaining between clashing elements, and detailed spectral resonant control. None of these maneuvers introduced any audible artifacts in the final summed audio.

I didn’t want to rush into offering this as an official service in the studio until I had tested it on a few more projects. Now, after working on more than a dozen, I can confidently say that FM Reconstruct is a powerful option for sonically resurrecting any of your lost projects - provided you still have the master file. For more details, check out the Rates section on my website.

Disclaimer: Stem-splitting technology still isn’t perfect for using individual stems in completely new productions. The artifacts become noticeable when any of the stems are used in isolation, outside the context of their original mix.

The studio has recently undergone a major update, including improvements to room acoustics, workflow gadgets and metering tools, as well as upgrades to the listening converters, analog summing, monitor room control, and digitally controlled analog bus compressor. However, the most significant upgrade is the reconfiguration of the sound system, which now includes the ATC SCM45A Pro and Auratone 5c active Sound Cubes, in addition to the existing Adam A77X system and Audeze LCDX headphones. Recently, I've been learning and utilizing SketchUp and AutoCAD to create precise 3D plans for the new studio layout, which are also displayed in the accompanying images. After two months of adjustments, calibrations, and experience in the new listening environment, here are some conclusions and impressions.

Working on the Adams in a partially treated room was always conducive to the creative process and vibe, but not fully sufficient for critical listening and discerning the nuances of mixdowns or individual sounds. This is where the Audeze LCDX headphones were always helpful. For mixing and mastering, it was always a 20/80 split between speakers and headphones. After the full room acoustic treatment with Anthill Audio, the ratio shifted almost to 50/50, with a significant improvement, but still slightly more in favour of headphones. During this period, the ATCs had arrived but were waiting patiently for robust speaker stands made of steel and filled with sand for the best resonance control and the ability to carry 50 kilos each for both Adam and ATC. The stands are inspired by Sound Anchor ADMID stands and were custom made by Davor Vrdoljak aka Venom, who not only met the main needs of the stands but also executed the option of height and angle rotation adjustment for each speaker. The next step was mounting the speaker "fortress" and finding the best position for the flattest possible response. Once we found the sweet spot, we tried to correct the minor bumps and dips with Sonarworks, but the difference was so small that I decided to continue working without corrections and avoid any additional phase shifts. It was cathartic to realise that the ATCs were well-suited to my room, and vice versa. This was a highly engaging and informative experience led by Goran Egić from Anthill Audio.

During the process, I listened to the ATCs but withheld judgment until everything was set up, and then it finally happened. I had many expectations, but they were mainly assumptions and somewhat abstract when, for example, trying to imagine even better transient response than in the Audeze headphones. I hadn't had much experience listening to this level of studio sound reproduction before. What shocked me initially was the mid-range representation of the sound picture that allows the effortless hearing and picking out frequencies inside the vocals or snare drum in the room. The mid-range was so obvious that it took over the first place in my first impressions. I understand that the ATC mid driver is the best out there with lowest distortion possible, but haven't expected this level of precision in the room when it comes to sound surgery.

After a little while, my attention turned to the low end, and there I experienced a slight paradox of underwhelming and overwhelming impressions. At first, it felt like I needed more bottom, but at the same time, I actually heard the sub notes and how they blended with the kick fundamental, and this was also something I could only grasp precisely inside headphones before. It was just right - tight and not hyped or blurry. I could talk more about low-end problems in general, but I'll just say that the ATCs handled it very well. In the future, integrating an ATC Sub is an option to extend the lowest octave and offer the full spectrum/club experience in the studio, but it's not crucial for the low-end adjustments for now.

The top end was my next focus, and this area, along with mid-highs, shares the guilt for ear attack and fatigue. After two months, I can confirm that the ATC tweeter is non-intrusive but an honest and assertive friend when it comes to catching ugly vocal sibilants or harsh "shurikens" from hats, snares, and other instruments. It just shows the problems without extra distortion in comparison to the slightly aggressive ribbon tweeter of the Adams.

While switching focus through the low, mid, and high-end, despite the ongoing awareness of the full spectrum, I then fully zoomed out and listened to the full spectrum. This, along with the impression of how surgical yet pleasant-sounding the ATCs are, led to a realisation of what's going on in the stereo field. The width, depth, and phantom centre of sound blew me away. This is one thing that headphones cannot reproduce realistically, even with cross-talk plugins. The soundstage experience of the ATCs includes their transient response quality, which allows you to really hear the start and the precise sustain/length of sounds and their position in the field. I can now fully rely on speakers for Mid/Side EQing and the width and length of reverbs and other spatial effects. It's amazing. This is surely due to the quality of ATCs' drivers and boxes, but they are also my first big (midfield) speakers I've worked on, so the whole impression is that everything is bigger and can be louder without extra distortions. I am immersed and hugged by the sound and feel like I'm inside the song, fixing resonances and clearly hearing what my analog gear is doing to the sound as well.

After executing about 80 masters and working on many mixing projects over the last 2 months, I can confirm that the usage ratio of speakers versus headphones has shifted to 90/10 in favour of speakers.

The next addition to the FM Mastering sound system was swapping the small A3X Adams for a new edition of the active pair of the original Auratone Sound Cubes, which are placed closely on the desk with a slight upward rotation. This is a nice addition for checking how the sound acts on mini reproduction systems like laptops or mobile phones, and so far, it works like a charm. Vocal ringings and other mid-range problems are nicely presented on these. They are also pretty good for checking how the low and low mids act on small systems.

The ATCs, Adams and Auratones are all plugged into the Dangerous D-Box+ with updated converters and are now seamlessly switchable with dedicated ABC buttons. Bringing ATCs in the house leaves no real need for any other speakers, but I decided to keep the Adams to have an option of listening to a different sounding system with a bit more distorted and smeared sound. Next to the ATCs, the beloved Adams now feel a bit more like a HIFI system, pulling back the midrange and acting more as a "smiley face" curve. The Adams' reproduction now leans more towards what people are accustomed to hearing outside of studios. The upgraded FM Mastering listening configuration of ATC + Adam + Auratone + Audeze is a major step up in my ever-evolving setup and has injected a lot of new confidence during mixing and mastering decision-making.