Hi HN, long time lurker Sam here. I run a small, bootstrapped audio software company called Musik Hack that this year started podcasting with artists and engineers in the industry. We took a left turn and did a Music AI roundtable with people in the startup (and traditional) space that I wanted to share here, because I think it's a very thoughtful examination of the excitement and ire people have about AI music right now.
With me were Zoe Wrenn of Tamber (congrats on the funding round!), Spencer Salazar of Output (one of the big startup players in the space), and two devs/architects from the gaming and audio world, respectively, Aaron McLeran and Eyal Amir. Hope it sparks some thoughts or ideas for the HN crowd!
I have an Audio Developer Conference talk about this topic if you care to follow the history of it. I have softened my stance a bit on the criticism of the 90’s (yeah, people were using lookahead limiting over exuberantly because of its newness) but the meat of the talk may be of interest anyway.
I enjoy this take. Funding something is not the same as creating it. The Medicis were not artists, Michelangelo, Botticelli, Raphael, etc were.
You might not be a creator, but you could make an argument for being an executive producer.
But then, if working with an artist is reduced to talking at a computer, people seem to forget that whatever output they get is equally obtainable to everyone and therefore immediately uninteresting, unless the art is engaging the audience only in what could already be described using language, rather than the medium itself. In other words, you might ask for something different, but that ask is all you are expressing, nothing is expressed through the medium, which is the job of the artist you have replaced. It is simply generated to literally match the words. Want to stand out? Well, looks like you’ll have to find somebody to put in the work…
That being said, you can always construct from parts. Building a set of sounds from suno asks and using them like samples doesn’t seem that different from crate digging, and I’d never say Madlib isn’t an artist.
A lot of what these AI tools do would classically be considered trying to “fix” a subpar mix, rather than just mastering. This is often not a super transparent change to the sound. For example, right up front, the first thing Logic’s mastering assistant does is put a full spectrum, highly detailed EQ curve over the track, ostensibly to hit some genre-dependent targets. Then it brags to you about it by graphing out the curve. This is not something many professional mastering engineers would do. Their job often consists of taking all the creative decisions made up to this point in the process and make subtle changes that bring them all out, much like adding the right amount of salt to a dish when cooking or polishing a car that’s already been detailed. They need to do this while making the track loud enough to compete in the consumer market, without crushing the dynamics or adding uncomfortable harshness to the sound. Generally, I think these AI products need to do less, not more, to get better results.
That said, this over-zealous AI modification isn’t always a bad thing, as the democratization of music recording has ushered in an era where many more people of various aptitudes are mixing music! But it does mean that AI-based mastering solutions run the risk of ossifying the “sound” of music genres, and while they can gussy up a poor mix, they often clobber a mix that’s already great. I don’t know of many major label albums where any of the tracks were mastered with an AI-based solution, although Ozone is in extensive use in the industry for its non-AI features.
Disclaimer: I make a non-AI mastering audio plugin called Master Plan.
This does work, but the sound of it isn’t super pleasing and it limits your filter design options. Checking out time domain IIR filters and methods for generating coefficients gets you flexibility and efficiency with no latency. I posted some links above but the RBJ Cookbook is a good place to start.
Big Star - “#1 Record”
Classic rock you could swear you’ve heard before but probably haven’t cause it had no radio play in its day.
Os Mutantes - “Os Mutantes”
What if The Beatles were Brazilian?
Chico Hamilton - “El Chico”
Just really good jazz that isn’t a household name but people into jazz know pretty well.
Porter Ricks - “Biokinetics”
Ambient electronic waterworks from the late 90’s
micronism - “inside a quiet mind”
excellent late 90’s electronic that was released without much fanfare
John Cale - “Paris 1919” and “Fear”
Just great rock/alt music from the early 70s. Especially check out the title track on Paris and Barracuda and You Know More Than I Know on Fear
Ben Johnston - “The Crossings: Ascent, String Quartet No. 4”
Accessible avant garde classical. Microtonal string quartet with a big, heart-filling and beautiful payoff.
Saint Heron
An awesome alt RnB/Soul compilation album from the naughties, I think curated by Beyoncé’s sister Solage, who also makes excellent music. Must be so hard having that kinda talent and be overshadowed by your sister so much.
Madvillain - “Madvillainy”
Not underrated, but unless you listen to boom bap indie rap, maybe hasn’t hit your radar.
Lil’ B - “I’m Gay”
The inventor of the term “based” drew major flack for this album title because he is not homosexual and meant “happy” but while confusing I don’t think that’s insulting at all. It’s the most accessible album I think he’s made and he is so prolific… man needs a greatest hits album because sifting through his work is normally a damn chore, so this album is a well compiled blessing
I could go on but this is my best I-can’t-sleep-tonight-here-is-a-top-of-mind list.
Just a quick nudge back on this: people in DSP would disagree with the assertion that nobody is going over big loops and using a virtual method on each element. We often have to process at least 88k elements per second in real time, through many many different processes. If any of those processes are defined using factories that spit out classes with polymorphic inheritance and virtual functions it certainly becomes an issue.
As a result some styles of writing code just don’t work for the audio thread at all, and we’d have to simply avoid or rewrite libraries written this way.
There are just some domains where standard practice for cleanliness is different because of your constraints.
I mean, it’s to the point we’ve got die hards in this industry who insist on putting all functions inlined in headers (not that I agree!)
This seems to be a corporate version of what I’ve witnessed: large groups of people getting together in vacation rentals or studios at writing camps and breaking off into small groups, each writing around a theme and passing ideas around fairly democratically, in a messy, creative, communal way. Then sharing and layering later. That leads to lots of contributors to one work of art, as they all reach for the best way to tell a story and fit it to a musical vibe that enhances it (or the other direction, it takes all kinds of approaches).
Never a whiteboard, so if that’s literally true I guess that environment would make me sad and disillusioned too, and I’m sorry your friend was dealing with it. But it’s certainly not the only model!
I agree with a lot of what you’re saying here. But there are two opposing points: if you doubt that removing aliasing would matter, then you can’t also hold that aliasing on purpose is an overlooked effect idea!
I have been doing tons of anti-aliasing and saturation DSP lately and agree you mostly should be picking a vibe and sticking to it as a DSP sound designer. However the hell that comes together in the math is irrelevant, as long as the minimal set of controls you provide to the musician are fun to tweak and you’re not delivering an airplane cockpit.
There is a lot of marketing blather about analog this legendary that but at the end of the day, a unique sound that is fun for musicians to “play” or intuitively useful for the tradespeople (mixers, mastering engineers, trackers) to use is all that matters!
It does. Oversampling is the classic way to reduce that problem, but some newer techniques to mitigate hard clipping aliasing (or any transfer function aliasing) with GitHub repo examples here:
With me were Zoe Wrenn of Tamber (congrats on the funding round!), Spencer Salazar of Output (one of the big startup players in the space), and two devs/architects from the gaming and audio world, respectively, Aaron McLeran and Eyal Amir. Hope it sparks some thoughts or ideas for the HN crowd!