Mastering engineers also seem to scrape at the barrel when it comes to explaining why it’s essential. Yes, you CAN do it yourself, but you will always be told that it’s best to pay the premium and get another set of ears to topcoat the song!
And now we have cloud-based, online mastering services that can master your song within minutes! What’s not to love about that?
Now…let’s see if AI can beat an experienced mastering engineer.
How The Experiment Was Conducted
The main reference for the overall vibe of the tune is pretty much anything from “Lullabies to Paralyze” by Queens of the Stone Age, a Joe Barresi production-especial.
There are quirky production choices obvious throughout and this needs to be reflected at every stage. It’s by no means a modern sounding production – we’re just not into that as a band.
The track was then mastered by these four “best” online mastering services:
I’ve also included the master submitted by my chosen human mastering engineer (Zac Zikis of Zikis Mastering), so you can be the judge of whether AI beats real ears in an actual mastering studio.
For consistency, I’ve gravitated towards similar settings for each of the platforms. It is worth noting that each platform has slightly different parameters available to tweak, so getting each platform to apply exactly the ‘same settings’ was a bit of a stretch.
I’ve rated each master as it comes and treat it as though I was receiving it for a release.
I’ve also assigned a score to each of them, with notes about what I liked and what I didn’t like.
Even though I’ve indicated the peak volume for each master after processing, I decided to normalize them to make the comparison fairer. This was only a very minor adjustment and I’m confident that it will give you a better idea of the sonic characteristics of each service.
Another note to add – I’ve used the lowest price plans necessary to get a ‘’HD’’ WAV master.
My Expectations Going Into This
I have never used AI technology to master a song so I can only guess. A real-life human mastering engineer has the ability to request technical changes based on their experience and overall judgment, which can help create a better end product.
When you start to squash a song to get modern-day loudness out of it, it can make elements of a song jump out more than the mix originally intended.
Surely AI isn’t going to care about this, It’ll just want to hit certain figures and hope that I like it, right?!
What I do know is that mastering can become tricky if songs are relatively dynamic. Certain parameters may need to be automated to preserve consistency from section-to-section. Can we really trust AI to have this sort of intuition?
The eMastered website has a nice, simple aesthetic and you’re immediately greeted with a big, bold heading that says “Master your track, instantly”. Apparently, eMastered was created by Grammy award-winning engineers, although it doesn’t say who these engineers are… Either way, if that’s true, then I guess I should know what it’s doing.
When you upload a mix, a series of vague terms such as “fixing audio defects”, “matching against billboard hits” and “optimizing for CD/Radio/Streaming playback” are displayed. Then something stood out to me (feel free to correct me if I’m wrong), the background image looks as though a guy has a pop filter placed at the top of a side-address condenser microphone…?! I hope this isn’t a visual metaphor for the mastering quality!
Once uploaded, the mastering interface is really nice. It’s simple and laid out in such a straightforward manner that even the most amateur of aspiring musicians/producers can navigate it and control the variables.
There is a real-time switching parameter that goes between the original mix vs the current master settings. This is great for direct A/B comparisons and there’s minimal lag when switching.
My only gripe is that the initial upload and the remastering process, once variables have been changed, takes a while. It feels noticeably sluggish to the point where you’re unsure whether it’s going to complete the task.
Once uploaded, I tried as many combinations of different variables as possible and in the end, I upped the compressor strength and also upped the mid-range EQ (there’s no specific frequency band option).
The first thing I noticed when analyzing the master was that there was no ‘top and tailing’ process to remove empty audio at the beginning and the end of the tune.
Maybe this is something that could be added as an additional feature?
It would also be good to add a feature to embed ISRC codes, I know this can be done via the distributor but every little bit counts, right?
During some critical playback analysis, the master sounded noticeably aggressive, more so than the others. On certain monitoring sources (mainly my Sony MDR-7506 headphones), the compression seemed to make the low-end pump, not that I seemed to mind from a listener perspective.
This pump extended the aggressive vibe of the tune. I am more than happy for a master to take the vibe into consideration so that the character is emphasized from all dimensions, not just from the instrument/tone sources.
LANDR is a user-friendly service that is probably the most well known AI mastering service.
Mastering is only one of the services that LANDR offers, so you will need to navigate to the Mastering section at the top of their site to find this.
Uploading your mix is as easy as it could be, and once you do this you get to hear a 30-second real-time comparison that is great for direct A/B comparisons. There is virtually no lag when switching, either.
It would have been good to be able to compare the whole tune as standard (as was the case vs eMastered). The song I uploaded was reasonably dynamic between sections, which was emphasized during mixing, and only being able to compare 30 seconds means I was second-guessing what other areas of the song will sound like.
I was hoping to be able to control more basic functions of the master on the basic membership package, but this is only available by upgrading. The one variable I do have control over is the “mastering intensity” which I assume is another way of saying “loudness”.
Like eMastered, you must remaster the file if you change mastering parameters, which doesn’t take too long but you have to be patient.
Again, there is no top and tailing present, and also no option to embed metadata.
I thought that eMastered was a long process, but CloudBounce was about to take me on a real journey!
This was also when I realized that each of the mastering platforms interpreted the mix much differently to one another, and it was proving difficult to even get them to sound alike. My main aim was to get the best master I could from each platform, but I just didn’t expect them to behave/sound so different.
The interface of dB CloudBounce is very much like eMastered. You upload and master your song in a virtually identical fashion, although presets that are genre dependent are readily available for you to choose, making the process even more effortless.
However, I decided that the presets weren’t for me. As a band we made some rather quirky choices, some we wish we had time to go back and change, and the best way to get the song to translate well through the master was to control the independent variables individually, as I interpreted them.
This particular master wasn’t my favorite, I really struggled to try and get a sound that I thought worked well, although the low-frequency detail was superior with this service.
The bass guitar maintains a strong fundamental throughout, without overwhelming other elements that share the same frequency range.
Another really annoying thing it did was warp the stereo representation of the guitars, the left channel was almost centered, and the only way around it was to up the stereo width mastering variable (and it the image still doesn’t sit flush as it did in the mix).
The only thing I could think of that could be causing this is some sort of phase manipulation/ correction?
Also, there was no top and tail… I’m seeing a theme here…
-10 LUFS (long)
Best bass detail.
Price – $9.90 per master with plans starting from $19.90 per month.
It’s also the cheapest by far, as cheap as 2 free masters per month! To download a hi-res master it’s only $6 (which lasts for a month) or $25 for an annual premium pass.
It’s not particularly quick. The initial upload process was average, but the mastering process was very slow. I feel less inclined to complain due to the price though!
If you’re using the basic package, you get the option to compare the entire original file to the mastered file, with the ability to adjust mastering intensity (again, I think this refers to loudness?). To adjust EQ you need to upgrade to the premium package, but for the price, it’s a no-brainer.
It actually sounds far better than you think considering the price point. It has a very modern sounding mastering style, fairly scooped with lots of shine on the higher frequencies.
This actually helped improve the shoddy snare! It’s not my preferred style of master for this style of music but I actually liked listening to its interoperation of how the song should competitively translate. I think with some small improvements to the user interface, MajorDecibel could be a top contender.
-11 LUFS (long)
Most modern and bright.
Price – Free with upgrade packages starting from $6 per month.
Zac from Zikis Mastering is a professional mastering engineer from Birmingham, England, with a great reputation for quality and service. I’ve known Zac since approximately 2016 and after I heard his work, he immediately became my go-to mastering engineer. He’s since mastered the majority of the music I’ve produced or engineered for others.
Due to the virus, our whole strategy had to change and we had to deliver the final mixes on a weekly basis with a very tight deadline, which Zac was able to cater for. I made sure that I left enough time to make potential revisions on the mixes, based on Zac’s recommendations to get a better master.
If there were any other problems, which there were in this particular song, he was able to change his strategy to try and amend the problems, whilst maintaining a coherent quality throughout the entire EP.
For this song, the ping-pong delay effect on the verse vocals are the most problematic element, and I wouldn’t expect an AI to fix this without affecting other mix elements. The first master made them cut through more than originally intended. This was then corrected as much as possible without physically being able to amend the mix. With that being said, this master has done the mix plenty of justice and it translates very well on all playback systems.
You can tell that lots of attention was paid to the stereo image and overall dynamics of the song. The only way that I could possibly make this master better is to fix the problems on my end in the mix and recording. So, in case it’s not clear, Zac passed with flying colors!
-13 LUFS (Integrated Loudness)
Retains the greatest integrity of the original mix.
Price – From £75
Score – 4.8/5
Conclusion & Final Thoughts
-11 LUFS (long)
-13 LUFS (long)
-10 LUFS (long)
-11 LUFS (long)
Modern & bright
From $9.90 per track
To cut it short, I award eMastered as the winner out of the AI platforms.
The aggressive quality of the master gave it extra stank, and with my listener hat on it just worked. I did try to replicate this with the other mastering platforms but they didn’t translate the same.
To be fair to the AI platforms, they are probably catered towards a more commercial market and I would love to attempt this comparison again with different genres and different mixes, music with less quirky production choices, or a more straightforward composition/production style.
So did eMastered do enough to beat the mastering engineer? Not for me. Is it a useable master?
Absolutely, and I would say all the AI masters in this test honestly sound great.
The thing that makes the human mastering engineer the winner in this case is the attention to detail.
The dynamics are preserved in a most natural way, even between sections of the arrangement. The mastering chain was obviously automated with a keen ear to cater for this. It’s the best representation of the mix, without heavily altering the tonality and/or balance of the tune.
As it currently stands, only human interaction can preserve these qualities by adapting skills and toolsets.
A mastering engineer does cost more than an AI platform though, so you need to decide whether your music is worth the additional attention to detail.
The behavior of the AI mastering platforms could make them better suited to different genres/mixes and how you want them to translate. So I’d encourage you to try out each service to see which one suits your style of production best.
If you want a bright and modern master, you may want to try MajorDecibel first. If you want a natural sounding master, go with LANDR instead. And if you want that undeniable human touch, get in touch with Zac at Zikis Mastering…if he’s not too busy with my own tracks.
Which master did you think was the best? Let me know down in the comments below!