ABOUT PERFECTION
Or how AI does ‘good’ for us

⌈ “Perfection in art is an ideal that artists strive for, but it is subjective and difficult to achieve. There is no single definition of perfection in art, as it depends on taste, era, and type of art. For some, perfection is the ideal representation of reality, for others it is the expression of emotions or innovation” - according to Google AI. ⌋

ONE OF MY MOST LISTENED TO albums recently is Frank Sinatra's The Sound of Blue Eyes, released this year. I don't even need to explain what it is – at first glance, it's clear that it's a collection of the artist's greatest hits. The title refers to Sinatra's nickname, “Ol' Blue Eyes,” given to him because of his striking blue eyes, a feature that was noticeable and captivating, especially at the beginning of his career as a teen idol.

And everything would be OK, everything would be as usual, there would be nothing to write about – just another remaster of popular songs, or maybe even just a compilation of older versions without any interference with the sound. It would be another ‘day in paradise’ for fans and the publisher – the former get something they already have, but in different packaging, while the latter release the same material for the 128th time, the costs of which have already been recouped many times over. Everyone wins.

If it weren't for the fact that the album was released by the unknown label Mouton Records, only in digital format, and with a subtitle. When we check out any of the streaming services, be it Tidal, Qobuz or Spotify, we find that in the last two years there have been more releases of this type of song collections popular in the 1940s and 1950s performers. These include: A grand selection of his beloved songs by CHET BAKER, HENRY MANCINI, SAM COKE, LOU RAWLS and DORIS DAY’S, Bonjour, Paris by AUDREY HEPBURN’S, and Once Upon The Page by PATTI PAGE.

The vast majority of these releases are from this year, although the first titles (Lou Rawls) appeared a year earlier. And again – that's fine, there are hundreds of such reissues. All the more so because, as it seems, all the albums in question are in the public domain, and the new publisher did not have to pay royalties for them, nor did it have to apply for a license from the original publisher, which in Sinatra's case at that time was Columbia Records, now part of Sony Music Entertainment, a subsidiary of Sony Corporation.

The fact that the material for this album comes from CD rips and not original tapes can be confirmed by the frequency spectrum posted on Steve Hoffmann's forum by a user named Cleantones; more → HERE. It is clear that the high-pass filter (“brickwall”) is set around 20 kHz, which means that the sampling frequency of the source material was 44.1 kHz. These recordings are available in 24-bit and 96 kHz because the mastering system worked with these values and it sounds better than if they were “cut” to the input signal settings.

So we are talking about an unauthorized release of old material that I have on dozens of CDs – why am I even mentioning it? The reason for my interest in this series is the sound we get. And it is, for the most part, perfect. Let me repeat: it is a perfect approximation of the vocals of the young Baker and Sinatra, and the wonderful female vocals of Day, Hepburn, and Page. I know them well, because the 1940s and 1950s are my point of reference, both in terms of the artistic and technical aspects of their recordings. Not that they are perfect, that's not what I mean at all. They are simply excellent, which is something completely different.

The new remasters are phenomenally rich, dense, and clean. The vocals have been extracted from the noise without being aggressively cut out. The background is clear, yet almost warm, not “shouting” as is usually the case. And above it all hovers a spirit of calm. It's as if someone has smoothed out all the rough edges that used to bother me when listening to the same material. Really – I've never heard anything like it before. Even very difficult recordings, such as the Hepburn album, which is not by a singer but an actress, sound quite good. And yet...

When you listen to a few of these ‘albums’, you will find that the advantages I mentioned earlier turn into disadvantages after a while. This is because all these recordings sound perfectly the same, almost identical – I am talking about the sound characteristics, of course. Is there density? Yes, there is, always. Is there saturation? Of course there is, but it's always there. What I mean is that they have been prepared in the same way, without taking into account the differences between them. It's as if someone had ‘locked’ the same settings in the mastering system and run the signal through an automatic machine.

And if I'm being honest, I think that's how it was and that it was possible to achieve such good results while homogenizing them, thanks to the emergence of new versions of mastering software supported by machine learning (so-called ‘artificial intelligence’). The most well-known and respected is iZotope's Ozone 11. It is a set of algorithms, of which the Master Assistant Ozone uses artificial intelligence to analyze and suggest equalizer, compression, and stereo imaging settings, speeding up the mastering process. Users can then fine-tune these suggestions or work from scratch, enjoying full control over their sound.

iZotope emphasizes that users retain full control over the mastering process, allowing for artistic adjustments and fine-tuning of the sound. This is something that will allow for spectacular results in the future, once people learn how to use the program. Here and now, however, as I see and hear it, the temptation to automate the process is huge, because the results are surprisingly good. This would not be the first case of AI being used in audio, as we have known for years about remixes of Beatles recordings, called ‘demixes’, made from monophonic or multi-track recordings, from which individual instruments and voices are extracted and remixed.

Peter Jackson, who directed the documentary trilogy The Beatles: Get Back, prepared a system allowing for such manipulations for this project back in 2017, together with his team. The results exceeded the expectations of not only the creators but also the audience – take a look at my review of The Beatles ’64 (Music from Disney+ Documentary) → HERE. However, the software was limited to this one band because ‘teaching’ the application the sound of each member was a long, difficult, and expensive process. Now anyone can do it at home with any album.

Which brings me to my main question: What are the goals of the creators of the new versions I started with? And although it's the middle of summer vacation, I ask this question with a heavy heart, because it's about more than just ‘another remaster’. Harnessing AI for this work makes sense; it's a handy tool that shortens many tedious processes where the human factor is not dominant. At the same time, however, it introduces a non-human aspect into the realm of music. While AI assistance is ubiquitous in new songs, and I'm fine with that, interpreting older recordings in this way makes me uncomfortable.

What exactly is a remaster, one might ask. I would probably answer that it is an attempt to deliver the original recording in the best possible form. What does that mean? – One would have to ask, that would bring out the hidden issue. What is the ‘standard’ against which we would compare these new versions? Hardcore purists will immediately answer, without batting an eye, that it is the first LP release. And that's fine, it's a very good point of reference. The only thing is that the vast majority of original recordings have their own problems, and AI promises to fix them – as in the ‘Restored Edition’ series of recordings.

Let's think about it in the same way that Janusz Sepioł does in his essay Architecture and Morality, published in 2024 by Austeria, when he asks about the restoration of historical monuments. Before reaching his conclusion, which we will discuss in a moment, he refers to Plato's triad: goodness, truth, and beauty, of which truth causes the most problems. This is because we do not know what is truly real and what is mediated and therefore altered. There is no need to refer to Plato here; it is enough to think about it.

Similarly, in audio, truth is a complex concept that cannot be explained by reference to a ‘live event’. A recording, even a live recording, is a new creation, an attempt to capture the inner ‘truth’ of music through technology and despite that technology. The fact that many believe in the identity of a live recording and the concert itself only testifies to the deep-rootedness of this belief and the skill of the sound engineers, not to its truthfulness.

No wonder then that with increasingly better tools at our disposal, we strive to go one step further in a direction that would allow us to better convey the intentions of the producer and musicians – intentions that are, of course, imagined. But if we don't have, because we don't have, an established, reliable point of reference, what should we be guided by – I say ‘we’, but I mean mastering engineers – when preparing new versions of recordings?

In this context, Sepioł refers to John Pawson's book Minimum, published in London in 1996. He says that the author wrote that it was intended to crystallize thoughts about simplicity, which should permeate architecture and art. Minimalism, Pawson believes, “can be defined as perfection, «which artifacts achieve when they can no longer be improved upon». This is the result, Sepioł adds, of avoiding what is beyond the essence.

In our case, it would be a matter of rendering the original with as little interference as possible, while ensuring that the overall effect is supervised by a person with good taste. Because, according to the author of Architecture..., the ultimate arbiter when making choices about ‘restoration’ should be good taste – and this applies to both architectural monuments and music. So, ultimately, it is always a matter of individual choice, and the end result depends on the person who makes it. Do you see the contradiction? Modern mastering is a very profound intervention, even though it is supposed to be ‘transparent’.

How can this be applied to mastering that is assisted, and often simply carried out, by AI? It is an algorithm prepared by humans, so it has all their biases ‘hard-coded’, including their taste. The only difference is that it is not used to extract the truth, but to average the results. Until ‘artificial intelligence’ becomes self-aware and creative, the effects of its work will consist of predicting the best possible next moves based on past events. So there is no question of creativity – everything has already been done.

That's why all the recordings in question sound similar and pleasant. I'm not complaining, I listen to them from time to time to de-stress. However, if I want to listen to something ‘real’, I reach for classic remasters and originals. I'm not a fanatic of the latter, and many new releases sound better, in my opinion. But I like their rawness; even the flaws are part of the music for me. When we eliminate all the ‘problems’, we may find that what remains is mush, food for children. But maybe that's what it's all about today, infantilizing the audience so they don't ask questions.

That is why listening to the tracks from ADAM CZERWIŃSKI's latest album was such a profound experience for me. He sent them to Julian Soja (Soyaton, KSS) and me in two versions. Both were mastered by Piotr Łukaszewski at Custom34 studio, but one was mastered using a new AI-assisted system, and the other using classic analog tools. The differences between them are enormous. At first, I was delighted with the ‘artificial’ version. It had everything I started this article with when describing the “Restored Edition” series. That is: fullness, density, and almost unbelievable smoothness. And a kind of depth that we long for in audio, describing it as a ‘black background’.

After some time, however, I realized that it all sounded the same. Beautiful, nice, but exactly the same. It lacked the inner truth offered by the analog version. Yes, the latter wasn't as pure and deep as the digital one. But for the purposes of a vinyl record, it seemed better to all of us, and Adam agrees, because it's better. As for the CD, the remastering with the new tool in Custom34 will bring us closer to the musical truth; the album will be released in the near future, as usual under the patronage of “High Fidelity” and with our ‘quality mark’.

There will be more and more remasters prepared using IA tools. The ease good results can be achieved with, i.e. reduced time and minimized costs, will be so attractive to publishers that all other methods will seem simply foolish. This will lead to the standardization of the remastering studio market. After some time, you will see that everything will sound the same. “AI” will choose what is most common, easiest for us to assimilate, and safe to perceive, rather than what is best or true.

Therefore, it is worth repeating after Sepiol that the human factor is the most important one and that its domain is ‘tragic choices’: ‘a series of dramatic choices of values’. And values are, for now, the domain of the living world. Let us not allow this to be taken away from us. He concludes his essay with a particularly powerful statement, quoting Josef Brodsky from an essey entitled Watermark, which also applies to the music market: “They babble about environmental protection, restoration of monuments, cultural heritage, and other such things. The goal of all this is one: constraint.” ●

WOJCIECH PACUŁA
Chief editor