“Our goal was never to make music that was good! Maybe someday we will, and people will say it’s good. But our goal is to make music that’s meaningful. That’s a different kind of thing…”
Alex Jae Mitchell is the CEO of Boomy, one of the newest startups exploring the idea of… well, I would say AI-generated music, but as he’ll explain later, he thinks that particular term is meaningless. Nevertheless, Boomy has developed artificial-intelligence technology that anyone can use to create songs, using its web-based interface.
You log in, choose one of two styles (‘Beats By You’ or ‘Relaxing Meditation’) then a filter (modern, high intensity, maximum thump, electronic, vintage, ambient or unlimited in the former case, and pulses, natural or electronic in the latter), with a third ‘Advanced’ style for people who want to tweak more variables.
Once chosen, it takes around 13 seconds for Boomy to create you a song. Play it, then either click ‘Try Again’ to get another one; ‘Save’ to save it to your collection, or ‘Edit’ to dive in and tweak its settings and arrangements. The promise is that the more you use Boomy, the better it learns your preferences, adapting its output accordingly.
Boomy is eight months old as a company, and launched its system in private beta in March. Since then, people have used it to create more than 110,000 songs (“around 0.12% of the world’s recorded music” as the counter on the site describes it). Boomy emerged from beta in late July.
The site is free to use, but offers an $8.99-a-month subscription for people who want to save and download as many songs as they like, but also to collect them into ‘albums’ – and even release those albums on streaming services like Spotify, Apple Music and Pandora – keeping 80% of any royalties that are paid out to Boomy by those services.
Mitchell founded the company in November 2018, having previously founded and Audiokite Research, a music-focused consumer-research startup which was acquired by distributor ReverbNation in 2016. Two years later, he left the latter company to explore his interest in AI music.
“One thought was that a lot of the research and efforts around AI music were ‘let’s take this style, rock, and make rock, or let’s take folk and make folk. I always thought that AI could be used to make things that were new, not just a best-guess copy of things that came before it,” he says.
“But another thought was about accessibility. If you could make this creative, and even make it reflective of an individual’s personal style over time, then you could open up this music-making thing to everybody.”
“I’ve had music education and a lot of people that I know have grown up in music, but in reality we’re a small group: there’s high barrier to entry for making music that we don’t necessarily see. That’s why there’s only 40,000 new songs going up to Spotify every day, versus 90 million photos going up to Instagram. Music-making is hard, and out-of-reach for most people!”
Mitchell is a trained musician – his education includes performing-arts institution The Juilliard School in New York – but his comments show a difference in perspective to some people in the music industry, who worry that 40,000 new songs every day on streaming services is too much, creating noise that talented artists struggle to rise above. Mitchell sees the number as evidence of a problem.
Boomy, like other firms before it, is thus in the ‘AI can democratise music-making’ school of startups. Who is it solving this problem for? That’s a work in progress.
The ‘Beats By You’ style on Boomy was created initially with grassroots hip-hop artists in mind for example: people who wanted to generate beats to rap over. However, during the beta it quickly became a word-of-mouth hit with gamers, who had a different purpose in mind.
“We saw gamer kids signing up to the beta and inviting each other like crazy. They were making songs to use for bragging on [messaging app] Discord. That notion of creating a whole EDM song just to say ‘Haha I beat you!’ to another player. It’s a use for music that wouldn’t ever have made sense in the past,” says Mitchell.
That was one lesson from the beta. Learning how to keep the site running when more than 30 people were using it at once another (“The whole thing would have come crashing down!”). A third lesson was that Boomy needed to be even simpler to use than its creators planned, having launched it with a creation-flow that involved setting tempo, intensity, instruments and genre influence.
“We thought we’d made the simplest music interface ever, with just four options. But people were confused: they had no idea what intensity meant, what tempo was. But to reach non-musicians you have to make it so simple to understand. To them, it was a little like if every time you opened your camera, it asked you for a lighting setting,” says Mitchell.
That’s what spurred the move towards ‘styles’ and ‘filters’ – the latter an intentional nod to Instagram – with the immediate effect of “all our stats going up by four times”. The original interface lives on with the ‘Advanced’ style (complete with the intensity up to 11, Spinal Tap-style) but the easier filters are front and centre.
How does it perform though? Sceptics will scoff that Boomy and similar startups couldn’t ever write a brilliant pop song, an amazing EDM banger or a properly-bouncing trap beat. That said, if you take 10 or 20 minutes to play with Boomy, testing the filters and rejecting a bunch of tracks, it does come up with some interesting sounds.
Can it really make ‘high-quality’ music though? This is where Mitchell makes the “our goal was never to make music that was good” comment, but rather “music that’s meaningful”. By which he means…
“People are going to stream this music not because it’s better than what’s coming out of labels or traditional musicians. That would be silly, to try to compete in that world. We want to make music that’s meaningful to a person,” he says.
“It’s about the context in which someone can start to apply meaning to their songs. Everything from the gamer using a beat to say ‘OMG look I won’ – that song means something to them now – to a yoga instructor creating an album of relaxation songs for their class.”
“That’s our goal: meaning for a user. I don’t think you’d see the kind of usage we’re seeing on Boomy already if people weren’t finding meaning. People are streaming their albums. One user has 50 monthly listeners on Spotify! Who are those people and why are they streaming it? Is it friends and family? We’re still exploring this stuff.”
The way Mitchell tells it, Boomy is less concerned with trying to create an AI that can generate a massive pop hit, and more concerned with creating an AI that can create songs with meaning to one person, and perhaps their circle of friends, family and (if gamers and other online creators pile in) fans too. The parallel is to what Instagram did for photography.
Even so, the fact that the company is actively helping users to put their Boomy-created music on streaming services does raise some sensitive (for the music industry anyway) questions about AI-generated music competing directly with human-created music. Which is when Mitchell challenges (with a smile, rather than anger) my use of those two phrases.
“AI-generated music sitting next to people-generated music? We’ve never thought of it that way. I don’t think there’s such a thing as ‘AI-generated music’. I don’t think that’s real! I don’t think there’s ever been a computer that on its own one day, with no input from a person, said ‘I’m going to make a song for myself…’” he says.
“The phrase ‘AI-generated’ doesn’t give enough credit to the user. I don’t think you’d call what Slash does ‘guitar-generated music’ or what Drake does ‘AutoTune-generated music’. It’s just music: it’s them as an artist.”
“The songs that people are making with Boomy, there are humans in every part of the process: humans on our side doing a ton of work to make this happen, and humans on the user side doing edits, adding vocals…”
Point taken: while it’s also true that hitting a ‘Try Again’ button until you get something you like doesn’t feel very creative, Boomy’s function for going in to to edit tracks after they’ve been created does add much more agency for its users in the process.
Like other people in the AI-music world, Mitchell compares unease at these kinds of systems with past controversies around the use of synthesizers, drum machines and AutoTune, among other technologies. “People were mad about big violas at one point!” he exclaims.
“The fear and discomfort around this [AI music] is really only going to come from somebody who believes that music-making should be reserved for a special elite class of person – that there’s this class of ‘musician’ and then there’s everybody else.”
“I believe that all people are musical, and when given the opportunity will want to express themselves through music. But if you don’t have that opportunity, not only are you shut out of the joy of making music in general, but you get left out of this incredible royalty pool [from the consumption of music in the streaming world].”
Isn’t that the most controversial point: that if anyone can create and release music through systems like Boomy’s, that it might end up taking royalties away from people who have made music their professional craft?
“The controversy only comes when you get to playing in that same sandbox. Who gets to play? Is there a kind of person who should be allowed on streaming, and another kind who shouldn’t?” says Mitchell. It’s certainly clear that Boomy’s decision to turn (in theory) anyone into an artist on Spotify for just $8.99 a month will bring these kinds of questions new bite, within the industry.
That feature also raises some thorny issues around copyright and authorship, though. Mitchell says that Boomy worked with a “great copyright attorney” to figure out its framework around distributing music to streaming services.
“The default, if you’re creating music on Boomy, is that Boomy as a company is going to own the right, although we give you the option to buy out every right, both composition and recording, for $19,” he says.
That would be an option for people who want to make this music available through their own choice of distributor. However, the $8.99 plan also enables them to operate through Boomy, which is where they get an 80% share of streaming royalties paid to the company for the music they’ve created using the system.
Mitchell stresses that Boomy is still exploring how best to handle all this, noting that for people who aren’t wanting to release their Boomy-made music, they essentially get a licence to use it for “most reasonable non-commercial and non-commercial uses”.
“If there is going to be a meaningful amount of music that’s created by people using automation tools, on streaming services in the future, then the royalties from that should go to the people who are creating it, and should be split fairly with the companies creating the automation systems behind it,” he says.
For now, Boomy is continuing to explore these kinds of issues; improving its product by adding new styles and filters, additional options like track length and even synthesised vocals; and seeing what the music industry makes of all this.
“If you go to any big label or publisher, you’ll find three types of people in the room with an opinion about this. The first are the people who go ‘That’s horrible, burn it with fire! We don’t understand it!’,” says Mitchell.
“The second are the people who see the world exactly the way we see it: who understand the future of creation is going to come from people, and that the barrier to creation is going to continue to lower – and that they need to react to that, to have systems in place and to be part of it.”
And the third kind of person? “Those are the people who go ‘the music’s kinda weird, I don’t really care!’” he adds. “Who gets proven right or wrong over time? It’ll be up to the market…”
Mitchell’s last point: he hopes that more investors fall into the second group of people, regarding AI music, suggesting that until this point the sector has been “way under-funded”. His pitch for that to change starts with the improvement in Apple’s iPhone camera over the last decade.
“You don’t need to know anything about photos: you press a button, now you have an amazing photo, which 10 years ago was just an okay photo. That’s because a huge amount of money has been invested in creating that camera, because it sells iPhones,” he says.
“Imagine if you had that same resource put into AI music. If you gave us $100m, you’d be able to have some very, very high-quality stuff, including good vocals. It’s just engineering time and effort: this isn’t a technology problem any more.”
“If there’s a sufficient amount of investment into this area, all of the technological challenges involved in creating really amazing music are solvable. And it’ll become a question of what kind of market there is on the other side of this thing. I think there’s a huge opportunity to be unlocked.”