It’s that time of the year again! The goose would like you to keep your personal comments to yourself, thanks. The old man’s hat is full of pennies (total value: 4m streams). And Music Ally is continuing our series of roundup posts for 2022.
Artificial intelligence (AI) has made great strides in recent years, and one area where it is beginning to have a significant impact is music. AI-generated music, also known as algorithmic or computational music, is created using algorithms and computational processes that mimic the human creative process. This technology has the potential to revolutionize the way we create and experience music, but it also raises important questions about the role of technology in the artistic process.
One of the biggest benefits of AI-generated music is that it can help reduce the time and effort required to create new pieces of music. Traditional music composition can be a labor-intensive process that involves long hours of practice and refinement. With AI, however, composers can use algorithms to quickly generate new musical ideas, allowing them to focus on refining and perfecting their compositions. This can help speed up the creative process and make it more efficient.
Another potential benefit of AI-generated music is that it can help bring new and unique musical styles and genres to the forefront. Because AI algorithms can process vast amounts of data and generate novel ideas, they can help create music that is unlike anything that has been heard before. This could lead to the emergence of entirely new genres of music and open up new creative possibilities for artists.
However, the use of AI in music also raises some important questions and concerns. One of the main concerns is that AI-generated music could potentially replace human musicians and composers, leading to job losses and a decline in the quality of music. Additionally, there are concerns that the use of AI in music could lead to a homogenization of musical styles, as algorithms are more likely to generate ideas that are similar to existing ones rather than truly original ones.
Overall, while AI-generated music has the potential to bring many benefits, it is important that its development and use is carefully considered and regulated to ensure that it does not have negative consequences for human musicians and the broader music industry.
Why is the text above in red italics? That’s so you can see which part of this article written by an AI. We prompted ChatGPT to “write a 300-word introduction to an article about AI-generated music, describing its benefits and its dangers” and that’s what it came up with. But we promise that everything from now on comes from a Music Ally human. Enjoy!
01 AI music startups were raising decent funding rounds
Music Ally has been writing about AI music startups since 2014, but for much of that time these companies tended to be bootstrapped. If they did raise funding, it tended to be in the low single-digit millions. That’s changing. LifeScore raised £11m in March from investors including Warner Music Group) while Endel raised $15m the following month.
Newer startups like Soundful ($3.8m in April) were also raising healthy seed-funding rounds, while backing of a different kind was seen when DAACI joined the Abbey Road Red incubator in November. Funding and expertise (whether it be from VCs or music industry mentors) will fuel the next round of growth and innovation for AI music.
02 A proper discussion about AI, training and copyright
There’s been a heated debate in recent months sparked by the growing crop of text-to-image AIs: the ones where you tell it what image you want by typing a description. The debate is about how these systems have been trained on the work of human visual artists, and the lack of permission, rights and payments for that.
This debate applies just as much to music, and in 2022 the industry woke up to it. A group of British industry bodies came out strongly against proposals to allow AI companies to train their systems on creative works, including music, “without the need for creators and rightsholders to provide permission”.
One problem here is that we don’t have an established licensing model for training AIs: that’s something that needs to be developed, so that the startups who want to do things properly can access commercial music. It won’t be simple, but it’s going to be important to figure out what the models may be.
We’ve seen some positive signs of good actors in the AI music world. Harmonai, which we’ll hear more about later in this piece, is training its system on public-domain / Creative Commons music, plus music submitted voluntarily by artists – an opt-in system in other words.
Musician Holly Herndon (more of her later too) and partner Mat Dryhurst have also been working on a standard called Source+, which would allow musicians, visual artists and writers to opt in or out of allowing their work to be used to train AIs.
It’s a good example of where the solutions to this challenge will come from: a meeting of minds between creatives who are interested in AI’s potential but also understand its risks, and AI startups and developers who want to do the right thing by those creators.
03 Some big music-streaming players are exploring AI music
Of all the AI music trends that might spook artists, it’s the sight of music streaming services exploring this technology. Actually, this has been happening for a while now, quietly. Spotify, for example, hired one of the key experts in AI music, François Pachet, back in 2017 to run an internal lab building music-making tools.
There were a few stories this year about what that lab has been up to, which we covered in our Spotify 2022 roundup earlier this month. Musician Benoit Carré used Spotify’s tools for his new album ‘Melancholia’ for example, then Spotify released an open source tool called Basic Pitch to help artists turn their ideas into MIDI audio to work with. A Forbes article talking about a work-in-progress tool to “take the harmony of a pop song from, say, Justin Bieber or Drake and combine it with the melody and rhythm of a fugue by Schubert or Bach, if that’s your thing” hinted at more to come (and possibly some new copyright wrangles to argue about).
Elsewhere, Apple bought one of the early AI music startups, AI Music (yes, it does what it says on the tin), although we suspected this was more about Apple’s own creative tools for music-making (GarageBand, Logic Pro) and video editing (Final Cut Pro) than Apple Music. Finally, Chinese music-streaming giant Tencent Music revealed that it had built its own line of AI singers using synthetic voice technology, and had launched more than 1,000 songs using them.
The fear in some quarters has always been that streaming services would ultimately see AI-generated music as an easy way to fill their mood playlists with tracks that didn’t require royalties to be paid to rightsholders (and thus human musicians). That hasn’t happened yet: the controversies around ‘fake artists’ on mood playlists turned out to be production musicians not AIs. But as creative AIs continue to improve, this question will resurface regularly.
04 As is one of the emerging big players in text-to-image AI
We mentioned Harmonai earlier on, and there’s a bigger picture around that project: the backing of Stability AI. That’s one of the big players in text-to-image AI with its Stable Diffusion system, so the fact that it’s operating in music too (with a Dance Diffusion tool released this year) is noteworthy.
Stability AI is well resourced thanks to a $10m funding round in October, and Stable Diffusion has already been used for at least one music-focused project: Video Killed the Radio Star… Diffusion, which transcribed lyrics from YouTube videos and turned them into prompts for Stable Diffusion images – which could then be used in videos for those songs. Which started another debate online about the implications for the writers of those lyrics…
Anyway, Stability AI is a significant player in the wider field of creative AI, and music is on its agenda. Oh, and it’s hired its own AI music guru too: Ed Newton-Rex, co-founder of Jukedeck, which was one of the first AI music startups to make waves (back in 2014) before later being acquired by TikTok’s parent company ByteDance. He’s now VP of product on the Harmonai team at Stability AI.
(Here’s a stray thought, spurred by that mention of TikTok. If text-to-music AIs become a thing – and that’s already being explored – social media apps could be one of the commercial uses. People might type in a description of the kind of music they want to accompany their latest video post, and the app could generate it for them. Given the excitement around licensing fees for commercial music used on these platforms, that might not be a welcome prospect for rightsholders…)
05 AI music is most interesting as a tool for human musicians
Music Ally is well aware of the controversies and sensitivities around creative AIs and human musicians, and we’ll continue to cover them – not least because some of the musicians who’ve been talking about the challenges (Holly Herndon and BT for example) are doing that from a position of knowledge: they’ve leaned in to the technology rather than kept their distance.
Still, we think there are a lot of positives to talk about, and it comes back to what human musicians can do with this technology. It’s a bit of a cliché to compare AI music systems to the synthesizer or the drum machine, but it still feels a relevant comparison. Both those technologies were controversial in their early days, with fears that they would destroy the livelihoods of human musicians. As it turned out, they were tools that inventive humans used to create new sounds and even genres (hello, electronic music and hip-hop). There’s still an argument that AI music will follow suit.
Some of the news that pinged our radar on this score in 2022 included music-making app BandLab launching a ‘SongStarter’ tool that used AI to generate beats, melodies and chord changes to help people get started with a song (see what they did there?) or just blow off some creative cobwebs. Spotify’s Basic Pitch was another example of this kind of technology.
Startup Staccato pitched its system as “an AI Lennon to your McCartney” with its ability to bounce off human songwriters and producers and help them figure out what might come next in a new song, Endel continued its series of collaborations with human musicians, with James Blake exploring what its system would do with his sounds and ideas. An idea we’ve voiced before – but it bears repeating – is that the most interesting things about AI music will be what humans do with it, not what it does to humans.
06 Boomy hit a 10m songs milestone (but what does that mean?)
We’ve covered US startup Boomy regularly since it emerged in 2019, and this year it hit a milestone. “Over 10 Million unique, original songs have been created with Boomy,” the company announced in November. Which by our calculations, meant that Boomy was being used to create around 14.8k new songs every day. In comparison, around 100k new commercial tracks are being uploaded to streaming services every day.
Creative AIs can make a LOT of music (or images, or videos, or text…) and Boomy’s milestone was a reminder that if we think the music streaming is a noisy world now, we could be in for quite the ride ahead if and when AI-generated music is being released at scale.
There are caveats. Creating a song on Boomy involves setting some options and pressing a button, listening to the result, and then either saving it or rejecting it depending on whether you like it. You could go to its website now and create a couple of albums worth of tracks in the next hour – but you might only save two or three of them. The more significant figure, then, might be how many of the 10m created tracks were deemed good enough to save.
Still, the sheer potential scale of AI music is something we should be thinking about more in the music industry: from collecting societies (still figuring out whether an AI can be an accredited songwriter, let alone how to deal with one that can spit out millions of songs a year) to streaming services, and the startups whose systems are creating and hosting all this new music.
07 Jolene, Jolene, take off them fabricated streams…
Yes, that’s a Dolly Parton / Kendrick Lamar lyrical mash-up, because this final point is about two stories that we wrote in early November. The first is a cover of Dolly’s ‘Jolene’ by Holly Herndon – or rather by her Holly+ ‘digital twin’ that she built in 2021. Holly+ provided the vocals, although the instrumentation came from humans.
The second is a cover of Kendrick Lamar’s ‘N95’ by an AI created by startup Splash: a showcase for its development of (as the company cheerfully described it (“an AI that can Rap!”). Again, humans provided the backing track.
Both of these tracks were a snapshot of how AI music technology is improving rapidly and being put to use. Neither is going to put Dolly or Kendrick out of work, but they were more a spur to think about how this technology might be used in interesting ways. Which brings us back to the necessity for any AI music project of considering the implications for musicians: what they might be able to do with this tech, and what it might mean for their work and careers.
While you’re here…
– Read our fuller primer on creative AIs and music, in which we ran through the various startups in the field, and considered some of those ethical questions.
– On a lighter note, read our conversation with ChatGPT about the music industry, tackling subjects including streaming royalties, NFTs and AI music itself.
– January’s NY:LON Connect conference we co-run with Music Biz has sold out of in-person tickets, but virtual tickets are still available. Check the lineup here!
This article was amended on 19 December to correct a mistake: Apple makes Logic Pro, not Pro Tools!
Get A Free Music Ally Account
Access unlimited News articles on the site, PLUS a trial of Music Ally’s subscriber-only services, including our industry-leading daily Bulletin email.