ArduinoMania

Music publishers have been on a spending spree in recent years, buying the catalogs and copyrights for songs of famous musicians at a frantic pace. Last December, Universal Music Publishing Group bought up Bob Dylan’s entire discography in a deal estimated at more than $300 million. Similarly, Stevie Nicks sold an 80 percent share of her works to Primary Wave Music for an estimated $100 million that same month. But as all this money changes hands for the industry’s biggest stars, one songwriting startup has plans to open the firehose of music royalties to the everyman.

“You see these huge deals, like the Bob Dylan deal with the publishing rights and all this money,” Alex Mitchell, co-founder and CEO of Boomy told Engadget. “It started with a recognition that most people are going to be left out of that and it caused us to have a conversation about equity in the music industry, 'how do we fairly remunerate artists, what's the role of labels,' there's just chaos happening in the music industry right now.”

Mitchell realized that one major obstacle keeping amateur musicians from becoming published musicians was a technological one. Setting up a home recording studio is no small task, and teaching oneself how to navigate the hyper-granular control schemes of professional-grade DAWs (digital audio workstations) like Ableton Live or Pro Tools can take months, if not years, to fully master. But what if you had an AI-based co-writer to handle the heavy technical lifting instead, similar to what Tik Tok and Instagram do for their creators?

“We really started looking at what it takes to draw creativity out of somebody, what kind of tool can you put in their hands — where there's so much of the process that's semi- or fully-automated — that they can just add their own layer of humanity to it.” What they came up with was Boomy.

“There's already AI being used in studios and in the music creation process,” Mitchell said. “A great example of this is Ozone auto-mastering. They have used artificial intelligence to be able to create great mixes, put great final polish on tracks, things like that.”

“So what we've done is we've taken a lot of those concepts and we've rewritten this stuff from the ground up,” he continued. “[It’s] less to think about how people usually make music, and more in the context of, if somebody doesn't have any skills at all, how fast can we get them making some stuff that they think is pretty cool?”

The web-based app is, essentially, a one-button music studio. Users can compose wholly original songs in around 5 to 10 minutes simply by clicking Create Song from the homepage, selecting the desired style of beat — whether that’s rap, lo-fi, experimental or “global grooves” — and then fiddling with the composition and mix until they’re satisfied. That song can then be uploaded to any of 40-plus streaming and social platforms where the song’s author can earn royalties based on the number of times their song is played.

Embedded below is a loopable, meditative jingle I put together during the course of my research. Despite my inherent lack of rhythm and general disinterest in music production, I found this to be a rather relaxing and enjoyable experience. After choosing the underlying beat and waiting a half-minute for the AI to generate a mix, the production process largely involved just shuffling icons around to adjust the composition and fiddling with dropdown menus to the instrument sets until I got something that I liked and think vaguely resembles the Konami menu screen music I grew up with. The entire process took less than 10 minutes.

Unlike recurrent neural network analysis models such as OpenAI or Google’s Magenta which, for example, can analyze Michael Jackson songs to be able to recreate the King of Pop’s signature sound, Boomy is not trained on copyrighted works. This is due in part because of the highly-segmented nature of copyright law, which varies drastically between nations and territories, but also because of the black box nature of such systems. If the infinite monkey theorem is any barometer, there is always a chance (albeit tiny) that a system trained on Michael Jackson might randomly spit out a perfect recreation of “Thriller.” And that’s very bad for the system’s designer.

“If I'm a music publisher and I own the rights to Michael Jackson,” Mitchell said. “I'm going to look at that model I'm gonna say ‘great, you know what, that's all mine’… if you're making a copy of somebody else's work, even if it's transformed, you're probably going to owe some publishing on that.’”

Instead, the team is taking a bottom up approach, leveraging previous experience in A&R research to train its AI in building beats and compositions from scratch. “We have some really advanced algorithms that are doing automatic mixing, deciding what sound should go together — what are the features of those sounds, how do those fit together, what is the perceived loudness rate of those sounds,” Mitchell explained.

Those features grew from a brute-force development approach — putting together various combinations of beats and compositions, then presenting them to beta testers. “In our first iteration of our model had a 98-percent rejection rate, but a 2-percent stay rate,” he continued. “And in that 2 percent, over millions of sessions, we started saying, ‘okay, here are groups of features that go well together.’”

Mitchell doesn’t view Boomy simply as a music creation tool, but as a means to achieve “​​the ideal world that we want to create," one which would allow creators anywhere on the planet to register themselves as a co-writer of their work alongside Boomy at their local publishing rights organization. However, because copyright law varies from country to country, Boomy has established an alternative way to ensure that songwriters get paid for their creative works.

“So what we're saying here is, a real world example would be, we just built a music studio, we filled it with great equipment, and spent millions of dollars building the studio,” Mitchell told Engadget. “You can come in and use it for free, make whatever you want, and on your way out, we're assigning you to our label, and we're going to give you an 80 percent rev share on everything we collect from what you made in the studio.”

“The IP vests with us,” he continued, noting that Boomy has been used to create more than 3 million songs to date, “which actually makes us, ironically, the largest record label in the world.” For users who are either already established musicians or otherwise want to obtain sole ownership of their songs, ”they can submit a rights request, and we can basically either sell the copyright to them or come to some other arrangement.”

While Mitchell could not share exact figures with Engadget, he did estimate that in the two years since Boomy’s launch, the company has paid out “tens of thousands' ' of dollars in royalties to its user base.

Moving forward, Mitchell foresees Boomy’s UI to add more additional control features and composition inputs, “over the next several months, we're really gonna focus and double down on vocal, melody and top line,” he explained.

The company is also working on new methods to earn royalties for its users. “We’ve got a bunch of influencer groups lined up and we've been doing some stuff behind the scenes to place tracks into YouTube videos,” Mitchell continued. “If you're a creator, or if you've got a podcast, rather than go pay for music rights, why not get paid for the music that you're using?”