thereport ISSUE 425 | 25 NOVEMBER 2019

AIs on the prize

What will AI-generated music and machine-learning tools for creators really mean for our industry? 1

ISSUE 425 25.11.19 AI REPORT

ive years ago, music:)ally encountered “There is a growing realisation that our first startup creating music with one: this is inevitable. Two: that this can artificial-intelligence (AI) technology, be used positively. And three: that if this is when British firm Jukedeck won a inevitable, and if it can be used positively, Fpitch contest at the Le Web conference in then there’s almost an onus on potential Paris. partners to lean in and get involved, so that Drawing on decades of academic they have a seat at the table, and influence research into both machine-intelligence on where this goes,” says Drew Silverstein, and musicology, the company’s aims were CEO of Amper Music. initially modest: to help people create music “If I were a larger business focused on for use in YouTube videos and other digital- content creation – musical or non-musical media projects. – I would have at the top of my priority list Since then, Jukedeck has been joined ‘How do we engage with AI music?’ You can by a host of other startups exploring the partner, build, buy or do nothing. But the role that AI might play in music creation, worst thing you can do is to do nothing.” from fully ‘AI-created’ music to tools for One important point, before you read this human artists driven by machine-learning report: music-creation is just one aspect of and neural networks. Bigger technology how artificial-intelligence technology can be companies like Google, OpenAI and even used by the music industry. Facebook have also published research Machine-learning, for example, is and tools. well established as a tech that can Any kind of automation in music-making power better, more personalised music has historically been controversial. Past recommendations. Meanwhile startups like arguments about whether synthesizers or Musiio and FeedForward are deploying it for drum machines would put human musicians the purposes of analysing and tagging big out of work (or whether tracks using these catalogues of music, to solve the problem of tools even counted as ‘proper’ music) reflect AIs on the prize un-futureproofed metadata. current sensitivities around AI. What will AI-generated music and machine-learning music:)ally will continue to cover these Yet in 2019, music:)ally has spotted and other uses for AI technologies within our some constructive engagement with these tools for creators really mean for our industry? industry, but for this report, we’re isolating technologies. Artists like Holly Herndon the music creation part of the puzzle. and Yacht have talked about how they used robots somehow replacing us,” says Lydia from industry-backed accelerators and “Like all new things, people AI as a creative collaborator on their latest Gregory, CEO of FeedForward AI, and incubators (Techstars Music and Abbey overestimate its impact in the short term, albums – finally providing evidence for the previously head of growth at Jukedeck. Road Red in particular). and underestimate its impact in the long ‘AI won’t replace artists, it’ll be a creative “But in the generative-music space, artists “I would be surprised if there are term,” says Stephen Phillips, CEO of tool for them’ theory that the startups have are working with it, exploring how we can any executives at any significant music startup Popgun. always expounded. work with this technology to do new things, companies that are not aware of what’s This report is our attempt to address “There’s this connotation of ‘robots rather than just focusing on the existential, happening in this space. And I think there’s an some of those imbalances and spark more-

report making music’, with the stock images – nihilistic questions.” effort, in some cases, to turn that awareness constructive discussions about what this there’s one of a robot playing a trumpet! The industry is leaning in, too, nudged into action,” says Alex Mitchell, CEO of one of technology is capable of, by focusing in on They play in to these deep-rooted fears of by the engagement with AI startups the newest startups in this field, Boomy. eight talking points. :) the 2

ISSUE 425 25.11.19 AI REPORT

the BPI, Jukedeck’s CEO reveals that Amper Music – whose product is similar to its software has already been used to Jukedeck’s – raises $4m of funding from a The timeline of compose more than 500,000 tracks. group of VC firms. Jan 2017 Apr 2017 A veteran of generative music, Brian Eno, British startup AI Music joins the Abbey Road releases his new album ‘Reflection’ as an Red accelerator in London. Its technology app that “creates an endless and endlessly uses AI not to create original music, but to modern AI music changing version”. ‘shapeshift’ existing songs.

e say ‘modern’ deliberately: May 2017 the theoretical work behind Google shows off another new project: AI-created music and NSynth. It’s a ‘neural synthesizer’ capable music-making tools goes of creating new sounds and instruments for Wback decades, while Brian Eno coined musicians to play with. the related term ‘generative’ music in the 1990s, when he was making it, not just Jun 2017 talking about the idea of it. Australian startup Popgun comes out of Still, the recent history of AI music and stealth mode in a music:)ally interview its engagement with the music industry revealing its Alice AI, which like Google’s A.I. kicks off five years ago, and a music:)ally Duet can play piano with humans. bulletin story about the latest intriguing- sounding music-tech startup... Jul 2017 It emerges that the head of Sony CSL in Dec 2014 Paris (of ‘Daddy’s Car’ fame), Francois British startup Jukedeck wins the startup Jukedeck Pachet, has quietly joined Spotify to focus on contest at the high-profile Le Web ‘creator tools’. conference in Paris, although its service is still in ‘early-access’ mode. Sep 2016 Feb 2017 Aug 2017 Sony’s Computer Science Laboratories Google shows off a project called A.I. Duet, Musician Taryn Southern reveals an album Dec 2015 (CSL) in Paris unveils two songs – Daddy’s which gets people playing piano duets with called ‘I AM AI’ for which she used several Jukedeck officially launches, having raised Car and Mr Shadow – created by a an AI. It’s part of the tech giant’s Magenta different AI-music tools to create the seeds $2.6m of funding. People can use its collaboration of AI and humans. initiative. for songs. software to make five tracks a month for free before they have to pay. Oct 2016 Feb 2017 Feb 2018 A Twitter bot called LnH AI makes headlines AI-music startups Amper Music and Popgun music:)ally spots a new AI-music startup, Sep 2016 in the tech press by composing metal, blues are among the participants in the first Mubert, which has launched a mood- Google’s DeepMind AI subsidiary reveals or jazz songs on demand for people who Techstars Music accelerator, along with music iOS app based on an algorithm that that it’s been working on a technology called tweet it requests. adaptive-music firm Weav. “composes infinite music that never stops”.

report WaveNet, with capabilities including music creation. Nov 2016 Mar 2017 Feb 2018 At an event hosted by music:)ally and Hot on the heels of its Techstars news, Jukedeck collaborates with Korean music the 3

ISSUE 425 25.11.19 AI REPORT

company Enterarts on some tracks and a Sep 2018 with Amadeus Code: a tool to write songs for Feb 2019 live concert, which sees K-Pop songwriters Amazon’s startups investment program, musicians and non-musicians alike. AI Music (the company) releases its first and producers using its technology. Alexa Accelerator, adds startup Endel, which consumer app, Ossia, which shows off creates “personalised sound environments Jan 2019 how its technology works by creating new Feb 2018 to help you focus and relax”. We find out what Amper Music meant remixes of songs on the fly for listeners. Popgun announces a seed-round of funding by ‘enterprise’ – it strikes a deal for from Silicon Valley VC firm Khosla Ventures, Nov 2018 Tencent Music to deploy its B2B AI-music Mar 2019 as well as Techstars Ventures, promising to Popgun reveals that its tech has progressed composition technology for users of its QQ Another new startup, Boomy, opens up its “create awesome original pop songs”. a lot: it can now play guitar, bass and piano Music service. private beta. It’s a tool for people to have “and each one of those AIs can listen to one original music created for them, by inputting Mar 2018 another and play together” Feb 2019 certain parameters. Amper Music announces another $4m Abbey Road Red adds another AI-music funding round, led by Horizons Ventures – Nov 2018 startup, LifeScore, whose co-founder and Mar 2019 whose previous investments include Spotify German startup Melodrive launches the CTO is Tom Gruber – previously co-founder Google reveals its latest AI-music project: an and DeepMind. beta version of its AI-generating music of the startup that created Apple’s Siri voice AI-powered ‘doodle’ for its search homepage engine for games developers, capable assistant. celebrating Johann Sebastian Bach. Mar 2018 of creating music in real-time to match Japanese startup Amadeus Code launches gameplay. Feb 2019 Mar 2019 an ‘AI songwriting assistant’ app that’s Startup Ecrett Music launches its service Endel makes headlines for becoming the designed to create seeds of songs for Jan 2019 for video creators, using AI to create first AI ‘signed to a label deal’ – although songwriters, or just nudge them out of Startup WaveAI launches a mobile app background-music tracks – the same model that’s quickly rowed back on, since the writer’s block. called Alysia, which competes most directly as Jukedeck and (formerly) Amper. partnership in question is just a distribution deal with WMG. May 2018 Google Facebook reveals that a team within its AI Doodle Apr 2019 Research department has been working on AI-music collective Dadabots launches tech for “translating music across musical a YouTube channel streaming live, AI- instruments, genres and styles”. generated-in-real-time technical death metal, round the clock \m/ Sep 2018 Amadeus Code emerges from beta with a Apr 2019 new feature letting people choose a song OpenAI, the non-profit AI company backed from Spotify, and have the app compose by tech mogul Elon Musk, unveils MuseNet: new melodies based on their chord “a deep neural network that can generate progressions. 4-minute musical compositions”.

Sep 2018 May 2019 Amper Music reveals that it’s shutting down Artist Holly Herndon is in the press talking

report its beta AI music composer, with plans in about her new album ‘Proto’, for which the future for “focusing on our enterprise she built her own AI called ‘Spawn’ to products” instead. collaborate with. the 4

ISSUE 425 25.11.19 AI REPORT

May 2019 Sep 2019 Zen 8 Amper Music publishes some research Mubert launches an app called Zen 8: Music claiming that humans couldn’t really hear a for Stories, designed for people to use to difference between human-created and AI- quickly generate music to soundtrack their created stock-library music. ‘stories’ posts on Instagram and Snapchat.

May 2019 Oct 2019 Amadeus Code raises a $1.8m funding Benoit Carré (of ‘Daddy’s Car’ fame) releases round, promising to use the money to the ‘American Folk Songs’ EP from his increase its global footprint, as well as Skygge project, using AI tools developed at further developing its technology. Spotify to flesh out a cappella recordings from US folk singers Pete and Peggy Seeger Jun 2019 and Horton Barker. Melodrive launches a 24-hour live Twitch channel broadcasting AI-generated piano Oct 2019 music. ‘Twitch Plays Melodrive’ was created Endel follows in Melodrive’s footsteps by as part of a hackathon with Dadabots. launching a sleep-music channel on Twitch. The channel - ‘Insomnia’ - broadcasts “a Jun 2019 Newton-Rex is now director of Bytedance’s collaborated with an AI trained on all 82 24-hour stream of real-time, procedurally Endel releases its second batch of five AI Lab. songs in their back catalogue. generated sounds”. AI-generated albums to streaming services, although when music:)ally checks Jul 2019 Sep 2019 Oct 2019 its stats, the company only has 872 Spotify Boomy launches out of beta, with a Russian startup Fluent launches Amadeus Code launches a new offshoot listeners. potentially-disruptive feature: people using an adaptive-music app for iPhone, called Evoke Music: a catalogue of AI- it to create original songs can now release personalising tracks as people move around generated, royalty-free production music, Jul 2019 those tracks to streaming services. based on their motion, the local weather and with a free tier and two levels of paid Dadabots makes headlines again for a the time of day subscription for clients. quirky AI-music project: Outerhelios is a Aug 2019 24-7 neural-network livestream of free jazz… Sony CSL may have lost its boss to Spotify, Sep 2019 Nov 2019 from NASA’s Voyager 3 space probe. but it’s still exploring AI music: its latest Tencent Music CEO Cussion Pang says that Greek startup Accusonus raises $3.3m project being AI-generated kick-drum tracks. the company is thinking about how to use of funding for its business, which includes Jul 2019 AI-music tools in the future, and that it is the Rhythmiq software, a virtual instrument Popgun publishes its latest video showing Aug 2019 “pouring in a lot of resources” to understand for musicians including an ‘AI beat the progression of its musical AIs. “For the Endel collaborates with artist Toro y Moi the potential. assistant’ tool. past year, we have been teaching an AI to on an EP of tracks released as part of a sing,” reveals the company. branding deal with bottled-water brand Sep 2019 Nov 2019 Glaceau Smartwater. London startup Auxuman kicks off a Jean-Michel Jarre launches ‘EōN’, an iOS Jul 2019 campaign of releasing AI-generated music app that generates an endless stream of

report music:)ally breaks the news that Jukedeck Aug 2019 by a group of five avatar musicians: the music using an algorithm working with has been acquired by TikTok’s parent Art-pop group YACHT release their new latest example of ‘synthetic reality’ being beats, chords and melodies that Jarre company Bytedance, where its CEO Ed album ‘Chain Tripping’, for which they used for musical purposes. composed and recorded. :) the

6

ISSUE 425 25.11.19 AI REPORT If you need music for a YouTube video, that is a solved problem Talking Point 1: already” – Alex Mitchell, Boomy Not just a threat to production music

ne of the key arguments in this libraries (bar Epidemic Sound) haven’t report will be that AI technologies really focused on anyway. For now, if not have much more potential to be forever, the higher end of bigger-budget creative tools for human musi- ads, TV shows and films may stick with Ocians than they do to become a threat to human-made music. those musicians’ livelihoods. Still, it’s fair to wonder whether However, there will be areas where forward-thinking production libraries AI-generated music competes with human- will be investing in music-generating AI made music, and production music is one of technology, through acquisitions and the first sectors where this impact will be felt. partnership, sooner rather than later. They The pitch has generally been towards may even have more flexibility to do so the lower end of that market: for example, than other companies within the music to YouTubers looking for background music industry on account of owning all their that won’t earn them copyright strikes; to rights. A production-music company has corporate-video makers wanting quick and the potential to train an AI on its catalogue, easy soundtracks; and in some cases for Evoke Music assess the quality of original music it use in games, TV shows and independent creates as a result, and perhaps build a films with tight budgets. That model is evolving. Amadeus Code’s is ‘My YouTube channel is about sport, new revenue stream as a result. Alongside Jukedeck, Amper Music recently-launched Evoke Music is a fully- camaraderie, leadership and competition’ It’s clear, though, that production music was one of the first companies exploring fledged production-music library full of and it would understand that, and show is far from the extent of most AI-music this as a business model. Its CEO, Drew tracks created by the company’s AI. whatever we have available.” startups’ ambitions, in terms of where they Silverstein, says that the value is already “It uses the songs that the AI has At a certain level, the ability of an AI to make their money. proven. “This is no longer R&D stage. This composed, and makes them available for create large amounts of music – teamed “The demand for that is only ever going is a commercially-valuable, and viable, anybody to use, whether it be individuals with humans to pick out the best-quality to be as big as the demand for production and successful tool creating inordinate who want music for their Instagram tracks for a library, although Amadeus music. If you need music for a YouTube amounts of music every month for updates, or vloggers, influencers or games Code is planning to use AI here too, to video, that is a solved problem already,” commercial customers,” he says. developers,” says CEO Taishi Fukuyama. feed back data on what clients request suggests Alex Mitchell, CEO of startup In the past, the onus has been on the “Eventually this is going to get smarter: and choose into the creation system – will Boomy. “If you start with a very simple people who’d use the music to control the we’ll learn what kinds of keywords people provide stiff competition for traditional problem like ‘people need music for their generation of it: by inputting their desired are searching for, and recommend music libraries. YouTube videos’ then you’ll probably come

report mood, length and other attributes, then that’s in our database... You wouldn’t be Well, it will in the field of YouTubers and to simple solutions that you can build a listening to the output until they found forced to become a music supervisor that lower-cost video creators, which in truth business on, but you’ll always be limited by something relevant. you really aren’t. All you’d have to say is a market that many production-music the market size.” :) the 7

ISSUE 425 25.11.19 AI REPORT Talking Point 2: Artists are getting hands-on with AI

hen Music Ally first started writ- metaphor because it felt like something that learning generated composition process” ing seriously about AI music in needed to be nurtured by the input that we that was trained on the band’s own 82-song 2017, the idea that this was a were giving it.” back catalogue. creative tool for human musi- Another prominent artist with an AI- “It still couldn’t make a song just by Wcians rather than a replacement for them infused album in 2019 was American band pushing a button; it was not at all an easy was floated by a number of people working YACHT, with their August LP ‘Chain Tripping’. or fun flow to work through,” said Evans’ in the field. At that point, it was mainly a They used what the group’s Claire L. Evans bandmate Jona Bechtolt. “So after three theory – actual evidence of artists collabo- described to Ars Technica as “a machine- days, we were like, ‘OK, I think we have rating with AI was thinner on the ground. French musician Benoit Carré’s project with Sony CSL in 2016; Taryn Southern’s ‘I AM AI’ album in 2017 and Jukedeck’s EōN and on collaboration with K-Pop producers Enterarts Jean-Michel Jarre’s ‘infinite music app’ in early 2018 were prominent examples, as was Brian Eno’s longstanding interest in It’s not an AI generating the music for Jean-Michel Jarre’s new app EōN. Jarre recorded ‘generative’ music – as expressed in apps its beats, melodies and chords himself. However, the app’s algorithm adapts these stems like Bloom and Scape. into a theoretically-endless stream of music. 2019, however, has been a breakthrough Jarre released the app in November 2019 as a paid download on Apple’s App Store, year in terms of prominent human/AI charging fans $8.99. He worked with Sony CSL researcher Alexis André as well as music- projects. One example is American artist apps developer Bleass on the project. Holly Herndon, who co-developed an AI “A lot of people are very pessimistic about our future – about a kind of dystopian (called ‘Spawn’) as a creative collaborator for I don’t see Spawn as a future regarding AI and saying that AI is going to be able to produce new and original her third album ‘Proto’, training it on her own content and wondering what’s going to happen to artists,” said Jarre at a press confer- human baby. I see Spawn as an voice as well as that of other people ence to launch the app at Lisbon’s Web Summit conference. “I don’t want to recreate music; I want to artificial intelligence baby He was keen to stress its innovation. “The app combines man and machine as the find a new sound and a new aesthetic. The – Holly Herndon, artist algorithm weaves together original material – beats, melodies and textures. It’s different major difference is that we see Spawn as an from a generative music app. It’s not a linear and ambient music type of app,” he said. ensemble member, rather than a composer. and can sing anything through it,” she told “I always dreamed about giving every listener in my audience the possibility of hav- Even if she’s improvising, as performers do, The Fader the same month. “I’m careful not ing a unique experience. When you start the app, you will have your own your own

report she’s not writing the piece. I want to write the to be too anthropomorphic with it. I don’t see experience and your own music that will never happen again. Every second of the app music!” Herndon told the Guardian in May. Spawn as a human baby. I see Spawn as an experience is creating some music you will never be able to recreate.” “I have a computer that knows my voice artificial intelligence baby. We use the baby the 8 Artists collaborating with technology opens up new ideas, ISSUE 425 new windows of opportunity, and new business models 25.11.19 AI REPORT – Isabel Garvey, Abbey Road Brands, artists YACHT Scored and Vochlea. and AI music Skygge Studios boss Isabel Garvey says that putting the tech Go to an event like the Cannes Lions, and into the hands of artists, you’ll see how excited brands and their producers and engineers agencies are about the idea of AI creativ- has been a key task. ity. It’s no surprise, then, that we’ve seen “We had to face the fear some musical projects in 2019 tapping and understand what this into this enthusiasm. technology and the power In August, startup Endel teamed of AI could potentially up with artist Toro y Moi to create four deliver. We need to map AI-infused tracks, each with a different out what the studio of guest vocalist. The ‘Smartbeats’ EP was 2050 is going to look like. released on streaming services, but the If this technology is going tracks were also made available through to be a serious part of how Endel’s website and app as “functional enough stuff.’ By that point we had a few The results were released in early October an artist records, and how they deliver their soundscapes”. The commercial partner for thousand clips between two- and 16-bars, as an EP, complete with a live showcase in vision in the future, we’re going to look like all this was bottled-water brand Glaceau and we just had to call it quits at some London. a dusty museum if we don’t move with the Smartwater, which is part of the Coca- point.” Long-time AI-music supporters think times,” she says, pointing to grime rapper Cola family. YACHT then used those fragments these collaborations prove that the Novelist’s experience testing Vochlea’s tech. Another brand-backed experiment to compose songs for the album, with a technology can be embraced by musicians, “You could see how quickly artists came in July, when LG worked with lyrics-generating algorithm too, setting rather than feared. collaborating with technology opens up new Amper Music and artist Betty Who. The themselves rules: that they could only use “I think we’re just starting to move on ideas, new windows of opportunity, and new Australian artist used Amper’s tech to cre- material created by the AI, removing words from the ‘will it take jobs?’ narrative, which is business models even. There was a total ate “the most motivational song ever” to and sounds if necessary, but never adding always a nonsensical place to start. I’ve been lightbulb moment for Novelist, as to what be performed at the LG-sponsored Color anything extra. banging on for as long as anyone would he could do with that microphone… And it Run event in Los Angeles. 2019 was also the year Carré debuted listen about these things being tools, not an ties back to the Beatles, and how they broke Amper’s system analysed hundreds his latest project, working again with former existential threat,” says Cliff Fluet, partner things deliberately, and made the studio an of motivational songs and playlists to Sony CSL boss François Pachet – now head at law firm Lewis Silkin and an advisor to instrument,” she says. identify shared ‘sound components’ like of Spotify’s Creator Technology Research a number of AI firms in his role as MD of Garvey recently caught a session at tempo, melody and chord progressions. Lab – on a project called ‘American Folk Eleven Advisory. the Web Summit conference where Major Who then worked with the results to cre- Songs’ under his (Carré’s) Skygge alias. “If you fast-forward five to ten years, it’s Lazer’s Jillionaire enthused about the ate the final track. The project started with old cappella going to be incredibly commonplace, and creative and collaborative potential of AI. recordings from American folk singers Pete just normal, that technologies like Amper “His whole message was ‘it’s going to Seeger, Peggy Seeger and Horton Barker, as are used as commonly as ProTools in the create new genres of music, it enhances bubbling up, there’s enough now for artists well as a new recording by Canadian singer creative process for music-making,” says human creativity, it takes down economic to experiment with, and to understand the Kyrie Kristmanson, and then using AI to Silverstein. barriers’,” she says. “I’ve been to a lot of capabilities.” “flesh out the melodies and the lyrics of the At Abbey Road Studios in London, the events where everyone’s a bit frightened and “The kneejerk reaction is giving way to a

report songs, enriching them with lush harmonies studios’ Abbey Road Red incubator has doom and gloom about AI. This was the first more detailed and nuanced understanding, and sounds generated by AI that have never included a number of startups working with time I’ve seen it turned on its head. Two to which then paves the way to openness and been heard before”. AI including AI Music, HumTap, LifeScore, three years in to these AI businesses really excitement,” is how Silverstein puts it. :) the 9

ISSUE 425 25.11.19 AI REPORT Talking Point 3: We need to think about training

opyright debates around AI music that training an AI falls under the ‘fair use’ the underlying copyright-protected material tend to focus on the output: for exception to copyright law. Other countries, it contains opted out or not! Elsewise, example, can an AI be a legal rights- like Japan, Singapore and China, have startups and businesses could inadvertently holder for a track, or does it have implemented similar exceptions for text and be infringing copyright,” explains Goossens. Cto be a human being? Those are important data mining to make sure that, in Goossens’ “But there is no incentive not to reserve! questions, but there are also issues around words, “copyright rules would not be able to By opting out, you create a system whereby the input – the data-sets that musical AIs are stand in the way of AI, when copies need to people might need to ask your permission, trained on. be made”. and you can charge for it. And if there’s “It’s something we should be asking Europe is a different kettle of fish. The an opportunity for music rightsholders to many more questions about,” says Sophie recently-adopted European Copyright charge versus an opportunity to give it for Goossens, counsel at law firm Reed Smith, Directive includes articles on this very issue. free? We both know where this is going!” who has been studying AI music and its legal There’s an exception for research Where does this leave AI-music startups? issues in depth. organisations, the scientific community As a generalisation, they should be okay Creative AI systems are trained using and cultural and heritage organisations: as training their AIs on copyrighted music in the a process of ‘text and data mining’ (or the directive is implemented by individual US, and in those other countries where text ‘machine reading’) that involves analysing a An act of copying is almost member states, they’ll be able to train and data mining has a fair-use exception. large set of data – music, in this case. And without needing permission from copyright As for Europe, unless they can rely on the one of the first things the system’s developer always necessary in order for holders. Startups and other [commercial] temporary copy exception, they may need to has to do as part of that training process is a machine to be able to read a businesses can not benefit from that carefully assess their position. make a copy of the data. training set exception. “This makes me sad, because I’m such a “With the state of the technology today, They can from another provision, supporter of startups here, but the answer – Sophie Goossens, Reed Smith an act of copying is almost always necessary but with a significant caveat. The is: until the situation is clarified, be careful in order for a machine to be able to read a the issue remains. directive’s exception for businesses allows when you train in Europe. Consider training training set. And this act of copying – unlike “How are you going to deal with that if rightsholders to opt out – to reserve the in the US or Singapore, for example, where reading, watching or listening [for a human you need to train your AI on large volumes rights concerning text and data mining of the legal context is really clear,” says musician] – is an act restricted by copyright,” of music? First, where do you find the million their content. Goossens. explains Goossens. pieces of music, legally? And second, if you “That caveat is significant and could Even in countries where fair-use should Startups may be able to claim that their need to make a copy of it in order to analyse potentially place a considerable burden on cover the training of musical AIs, it’s clear copy of the training set is ‘temporary’ – it, do you need to ask permission to each the shoulders of startups and businesses startups are taking a cautious approach.

report something covered by an existing copyright and every rightsholder?” who would arguably need to verify, each “There are no publicly-available, good and exception in Europe – but for anyone Do you? That depends where you’re doing time a training set needs to be copied in a large data-sets. A lot of the investment we’ve needing to keep a copy of the training set, the training. In the US, the general view is permanent fashion, whether rightsholders of done is building up our data set, because we the 10 We are certainly able to build data-sets on specific ISSUE 425 AI REPORT 25.11.19 pieces of music, or playlists, or any number of inputs – Drew Silverstein, Amper Music Elton’s Afterlife When Elton John announced his final tour in January 2018, the headlines went to a companies to explore the potential to mine virtual-reality film taking fans through key their back catalogue. Abbey Road’s Isabel Garvey strikes a moments from his career. However, John’s note of caution, explaining that sometimes partnership with creative agency Spinifex AI-music startups can underestimate the and tech startup Rival Theory could go well potential complexities of these partnerships. beyond that in the future. “At what point is that the artist’s The latter company creates “AI perso- composition, if the machine is blending nas based on real people”, with the aim in stuff, and where are the rights? There can John’s case to create a virtual Elton that be a bit of naivety, or underestimation of the will live on after the real Elton dies. Wired legal rights environment. It gets all the more reported on Spinifex CEO Ben Casey’s complex with AI. So startups need to have ambitions: “If he can somehow use artificial some empathy for that: it’s not figured out intelligence to unlock Elton’s songwriting yet,” she says. abilities, future artists could collaborate As Goossens has explained, AI startups have a freer hand to train their systems with virtual Elton in years to come.” don’t train on outside material or data,” says idea in their head at the moment,” says in countries like the US, but even there, Training the AI on Elton’s back catalogue Amper’s Silverstein. Goossens, adding that even within the partnerships with individual music “Our sample library is now the largest in world of AI startups there is surprisingly little could be part of that, and it’s a strategy companies or musicians have creative the world: it took us four years to go from awareness of the ‘rights-reserved’ element that other artists may explore. “Perhaps you potential. It’s something that Amper Music zero sampled instruments to 2,000, and to the European copyright directive. start with estates. Dead artists aren’t going has been thinking about already, but CEO less than another year to go from 2,000 to One challenge if music companies were to be making new music any time soon!” Drew Silverstein says that the legal issues over 9,000… From a legal and a copyright thinking of creating licensable training says Cliff Fluet. are never far from his mind. perspective, we believe we’re on very solid data-sets is the music industry’s rights “We are certainly able to build data- ground. Without any access to the outside structure. A major label might have a sets on specific pieces of music, or world, or any intent to create music inspired catalogue of recordings, but consider how to strike inventive, exploratory deals with playlists, or any number of inputs. And are by something else, we can create music free many publishers would have ownership (or AI-music startups, and possibly find a new, happy to do so in very specific use cases, of copyright concerns.” often part-ownership) of the compositions cost-effective way to create music that when the partner who would like us to do There is one question that arises for those tracks. Past major-label attempts unlocks new revenue streams as a result. that can show and prove that they have from all this. If musical AIs need to be to create ‘sandboxes’ of recordings for [non- For companies often thought to be most at all of the rights to the music that they’re trained on musical data-sets, but if – AI] startups to work with have run into this threat from music-generating AIs, it’s surely asking to be used as the input to Amper’s as Silverstein says – there is a lack of challenge. worth considering. model,” he says. publicly-available data-sets for startups That said, labels aren’t the only music Lewis Silkin’s Cliff Fluet thinks that “If someone says ‘This is my music, I to use, could this be an opportunity for rightsholders. Production-music companies, music publishers may also have an want you to build a digital version of me’ the music industry? Could rightsholders for example, may have a rights-ownership opportunity here: by taking works that and it’s clear that person has not only the create well-packaged data-sets and then model that would lend itself better to they wholly own, and commissioning legal ability to allow us to do that, but that license them (for a reasonable fee) for creating data-sets. new, original recordings that could then they’re willing to do that, then we can create

report startups to use? That’s an interesting prospect explored be used as part of a training data-set by descriptors that are really built around that Well, they could, but… “Not many people earlier in this report: that production- an AI startup. Or perhaps even putting external material… but only if there is a clear in the music industry seem to have that music companies may be in pole position specific songwriters in a room with these legal authority to allow us to do that.” :) the 11

ISSUE 425 25.11.19 AI REPORT Talking Point 4: AI can’t find the needles in its own haystack

nce it’s up and run- These “It’s just not possible right now: to say ning, a musical AI instruments know that’s a solved problem is just not credible can create music at how to play the to anyone who works in this space. a scale that dwarfs ‘Humans can’t predict whether this song piano! But they don’t Othe output of even the most- will be a hit, but my AI can’? Even people prolific human musicians. know if what they who don’t work in music AI, but work in Amper Music’s system can have done is good AI, would dismiss that as not possible,” he create a minute of music in says. – Stephen Phillips, Popgun about 5-6 seconds, while Music-generating algorithms need Boomy’s ETA for a new piece humans to curate their output, as well is 13 seconds. Left to their to work out which ones as to have a creative input into what own devices, these AIs could they are,” says Popgun’s they are producing. That’s a point that pump out a phenomenal Stephen Phillips. Boomy’s Alex Mitchell is keen for people to quantity of original music, “We’ve spent two years understand in the music industry. and due to its sheer scale, working on how we model “There is no such thing as AI music! within that would be some these instruments so they I think it’s a farce! Robots don’t make impressive quality. news. “There’s still the matter of being sound like humans are playing them. music: that’s just not a thing that Amadeus Code’s Taishi Fukuyama cites able to curate: things that are curating These instruments know how to play the happens. A lot of the conversation around a short story called ‘La biblioteca de Babel’ and discriminating. We and others will piano! But they don’t know if what they this is imagining a scenario in which there (The Library of Babel) published in 1941 continue to push the creative algorithms, have done is good. Right and good are is a fully-automated, sentient being – with by Argentinian author Jorge Luis Borges but the other, discriminative algorithms different things. They know what they’re no human intervention – that wakes up to illustrate this potential. The library will be very human.” playing is not out of tune – they’re trained one day and says ‘What I, as a sentient includes a book for every single possible An AI can create beautiful music, but on stuff that’s in tune – but they don’t computer, am going to do today is create combination of the letters of the alphabet it doesn’t know it’s beautiful. This is a have any clue if what they are playing is a song for myself’. There’s no such thing,” plus commas, spaces and full stops. challenge that AI-music startups are well original or interesting. That’s why we need he says. Most of the books are gibberish, but by aware of, too. humans to curate it.” “There’s no such thing as AI music! definition, the library also includes every “We haven’t tackled the core problem How long will this be true for? There are It’s just people making music. What single masterpiece that has or will ever be of quality or taste. I can go and jump on already a clutch of other startups exploring we’re doing is creating music automation written. our piano model and spit out a million whether AI is capable of spotting a hit purely systems… and those are created by “It’s the same for music. There’s only piano tracks, eight bars long in any by analysing its musical properties. But people. And then other people come in

report 12 notes. Set to a four-minute duration, different key, and within that million Phillips is unconvinced that this technology and make their own songs, mixing them an AI could write every single song,” says there’ll be one or two that are amazing. will be ready to be partnered with music- up. What you have is people making music Fukuyama, before delivering the bad What I don’t have is a computational way generating AIs in the near future. for people.” :) the 12

ISSUE 425 25.11.19 AI REPORT

AI shouldn’t just compose Talking Point 5: AI music can original music. It should sound completely different… sound weird (and it should do!) – Stephen Phillips, Popgun

e’ve established that AIs can ‘Move 37’ in the series’ second match, Sophie Goossens. “And some of the music create a LOT of music: Boomy made by AlphaGo, was a seemingly- was indeed strange: it had a really unusual only launched in March, and ridiculous move that no (sensible) human ring to it. It’s not what you would expect its users have already created player had ever tried. It also turned out to be a a human to write, unless it’s someone – Wnearly 250,000 songs. If you’ve heard some genius move that visibly stumped Sedol, and maybe Björk? – who was trying to make AI-generated music, it’ll only be a small frac- led directly to AlphaGo winning that game. music that sounded strange and bizarre!” tion of the material that’s out there. It sparks this question: what might Abbey Road’s Isabel Garvey agrees, not Even so, people are often happy to be the equivalent of ‘Move 37’ for AI- unapprovingly, and relates this AI-assisted jump to generalisations about the quality generated music? Something that initially project back to some famous human musical of machine-made music, based on what seems strange and alien, but which might experimentation. they’ve heard. And often, the generalisation actually point humans in a new direction “He’s taught the computer some rules is, specifically, that the music is bad. for music-making. of composition, and it’s churning out Sometimes, it is. “Often, you can be “AI shouldn’t just compose original rhythmically and tonally-accurate stuff, dazzled – or rather un-dazzled – by the music. It should sound completely different. but it doesn’t always fit very well. It can be fact that it’s using poor-quality stems. It should be pleasurable, but also ‘Holy hell! weird chord progressions, because it’s not You’ve got to look at the complexity and Like nothing I’ve ever heard before!’ We’re restrained by our rule base, and what we the sophistication of the underlying out something which to human ears only just starting to work on that,” says consider good. It comes up with something algorithms,” says Cliff Fluet, from law firm (subjective, admittedly) sounds ‘good’. Popgun’s Stephen Phillips. very new, and that’s where the creativity Lewis Silkin. “The question is always about “You’re going to start seeing the music “What happens if a lot of people say starts, even if some of it makes for difficult what stems you’re using. Some established get really consistent, and really good, a ‘that sounds weird, that doesn’t sound listening,” she says. stems libraries, strangely enough, haven’t lot faster than I think people necessarily right’ then?” says Amadeus Code’s Taishi “But that takes us back to something been willing to license to AI companies so realise. That’s going to be interesting: it’s Fukuyama. “We might have to say that this like ‘Tomorrow Never Knows’ by the far. That will change.” going to be months, not years, until it’ll only may be the statistical trend that the data Beatles, which was described at the “The more you see where the take one or two spins to get to a pretty cool is showing us: maybe it’s a new way of time [of its release, in 1966] as a mess technology starts and stops: it’s only as dance song, or a pretty cool trap beat,” showing us what pop songs are?” of a noise, and frightening. Tape loops, good as its inputs, and creative human says Boomy’s Alex Mitchell. On this score, the recent performance of backwards guitars, Lennon’s voice fed input is a necessity,” agrees Abbey Road’s This is not the limit of ambitions here, Benoit Carré’s ‘American Folk Songs’ project through a Leslie speaker. All sorts of crazy Isabel Garvey. however. One of the most famous moments in London sparked thoughts from several of stuff going on, and a lot of people didn’t Better stems (and data-sets – see in the recent history of AI came in 2016, the people interviewed for this report. like it. Now it’s been called one of the top earlier this report) will help, while the during a series of matches of Chinese “Some of the music was very difficult 200 songs in the world. It’s those moments

report systems themselves are improving boardgame Go between one of the best for a human voice to sing: the singer had of craziness, whether done by artists in the rapidly in both quality and consistency human players, Lee Sedol, and Google’s a struggle, in places, to sing what the AI studio, or by an AI, that could define the – the regularity with which they can turn AlphaGo AI. had written for her,” says Reed Smith’s next 200 years of music.” :) the 13

ISSUE 425 25.11.19 AI REPORT Talking Point 6: AI-generated music goes D2C

ntirely-AI-generated mu- DSPs may be wary of pursuing 70% of our users never made sic may be good for the this route, though. Lewis Silkin’s an original song, or thought about soundtracks of YouTube Cliff Fluet notes that 2017’s furore making an original song, before” videos, but people won’t around so-called ‘fake artists’ on Echoose to listen to it on its own, Spotify – actually production-music – Alex Mitchell, Boomy as an alternative to human-made artists creating mood-music under music… will they? Actually, perhaps pseudonyms – offers a glimpse of song, before,” says Boomy’s Alex Mitchell. they will: there are several contexts the potential controversy if that “People who are not musical, or who don’t in which AI-made music can already streaming service ever enlisted AI for know how to play music, are participating in be listened to – by choice – whether this purpose. the rights system. That’s what I find exciting.” through bespoke functional-music “They created a moral panic Again, at this stage Boomy-created apps or on the mainstream music- about that, which has meant there’s music isn’t giving the top human musicians streaming services. a moral panic about AI. It was a very on Spotify or Apple Music a run for their Some startups are straddling good hit job,” he says. money. And the prospect of ‘people who are both those sectors. Endel is a prime Another way AI-generated music not musical’ piling in to a streaming market example. Its launch product was is making its way onto commercial where more than 40,000 new tracks are a mobile app – and later Alexa streaming services is via Boomy, already uploaded to Spotify every single day skill – offering “personalised sound which allows people to compile the will raise the hackles of many people in the environments to help you focus, tracks they’ve made using its system music industry. relax and sleep”. It has since been into albums, and then release them Still: it may not be long before an AI can stretching into all kinds of other on the main DSPs under whatever do as good as (if not better) a job than a places, from an Apple Watch app artist name they choose. They get human at creating music that’ll send you to and 24-hour Twitch channel to in-flight for the future. 80% of the royalties for the recordings, while sleep, help you focus on that work task, or entertainment systems, retailers and even Where this will get really interesting Boomy gets 20% plus the songwriting credit provide a suitable accompaniment to your ‘mass meditations’. is the ‘personalised’ element, where an and publishing royalties. meditation session. Yet Endel has also styled its algorithm AI is generating music based on all kinds This is interesting, because it could This isn’t controversial, and human as a ‘recording artist’, releasing a series of contextual data from its user – from create a new class of ‘non-musician’ artists artists’ music will still be able to play a role of albums (through a distribution deal location and time of day through to their competing for attention in the streaming in these areas. But as more pipelines pop with Warner Music Group) to streaming heart rate and past listening preferences. world. up for AI-made music to make its way onto services and download stores. These aren’t It’s here that AI music makes more sense as “There is a growing catalogue of our commercial music services, there may be

report popular yet – Endel has just over 1,500 a standalone app – or a dedicated feature users’ albums on streaming services. And more arguments ahead – especially if, as the monthly listeners on Spotify at the time of within a streaming service – rather than as 70% of our users never made an original following pages will explain, teenagers get writing – but they are a marker laid down commercially-released albums. song, or thought about making an original hold of the means of production. :) the 14

ISSUE 425 25.11.19 AI REPORT Talking Point 7: “Instead of Taylor Swift pretending to be a 15 year-old, The real endgame for AI music? there’ll be an actual 15 year-old, talking about the things e’ve explored AI-generated music for functional purposes, 15-year-olds relate to most be it YouTube background mu- – Stephen Phillips, Popgun sic or relaxation soundscapes. WWe’ve looked at some of the ways this music year-old girls say to each other, so they can is making its way onto commercial stream- write songs for them. 13 year-old girls know ing and download services. And we’ve what they say to each other! Once they have talked about how AI technologies including the tools to make music that sounds great, machine-learning and neural networks can they’ll make music for each other, and it’ll be creative tools for professional musicians. sound incredibly genuine to them,” he says. What we haven’t yet talked about what “Instead of Taylor Swift pretending to be a is what might just turn out to be the most 15 year-old, there’ll be an actual 15 year-old, disruptive element to creative AI in the talking about the things 15 year-olds relate music sector: user-generated content. What to most… And I don’t think it does anything happens when tools for creating music but help music. labels will exist to harvest (using an AI) are put into the hands of tens the best and brightest of those kids and turn of millions of teenagers, in the same way them into pop stars. Whatever happens with that Instagram and TikTok put photography AI music, there’ll still be 20 super-engaging, and videography tools into their hands? charismatic people better than everybody This prospect is something that a lot of else at the top of the charts” AI-music startups are talking about, actually. The vision is of some kind of tool (likely Popgun’s Stephen Phillips, for example. become legitimate pop stars? Who’s going intelligence tools to create it. Phillips agrees. an app) that puts AI-generated music at “What’s the end-game for this? There to create that space where the next Billie “It’s us who give the meaning to the the fingertips of tens of millions of creative isn’t this place in the world where teenagers Eilish comes from? The current pop industry music: it’s just functional music if there’s no young people. Could the next TikTok-sized come together to make music for each other. is very few musicians controlled by three or human associated with this. Billie Eilish is youth craze be an AI-powered app for That place does not exist, and that’s nuts! four companies, played by a billion people. not a star because her stuff sounds so much making and sharing music? Or, indeed, That thing needs to exist, and it will exist. The new thing has to start with kids making better than angry teenage stuff we’ve heard could AI music be the next big feature for And getting the AI working is the price of music for each other and ignoring all of that before. She’s compelling as an artist,” he apps like TikTok? admission to build that thing,” he says. world. It’s still too hard, though, and AI has to says, before speculating about the potential It’s at this point that you remember that “Where’s the ‘pop stars on training be the answer for that.” impact of putting tools into the hands of TikTok’s parent company acquired AI-

report wheels’ place where they make music for This goes back to Alex Mitchell’s point more teenagers. music startup Jukedeck earlier this year, each other, release it and watch each other about ‘AI music’ actually being people- “Think about all these twentysomething and puts its CEO in charge of its new ‘AI pretend to be pop stars, but then go on to made music, even if they’re using artificial- hipster-songwriters in LA googling what 13 Lab’ in London. And also that Triller, an app the 15 “Do we want to give to someone who just pushes on a button ISSUE 425 25.11.19 AI REPORT the same copyright and same rights as to a professional musician? – Sophie Goossens, Reed Smith

similar to TikTok, recently acquired another be in the curation piece. “If you just have a idea that basically breaks the biggest rule because if lots of non-musicians are using AI-music startup, MXX – even if the latter place flooded with very-mediocre content, of all: that the music industry is driven by a AI tools to make music for one another, was focused on reshaping existing music it’s going to be a pretty awful place to limited supply [of music]” sometimes they may find that music to videos, rather than generating original be! How do you help the good stuff get Alex Mitchell voices similar sentiments. more interesting than the music made by tracks itself. through: what kind of algorithms do you “What’s the next hip-hop? What’s the next professionals. “As an independent, a company like need?” says Garvey. huge revolution in culture in the way this There may be other tensions too. “Do we Mashtraxx or Jukedeck might have 20 or 30 Perhaps the algorithms would be human: stuff gets made, and consumed? I think want to give to someone who just pushes clients. Or they can have one of the world’s a huge community of teenagers creating this is it. It’s a few years out: it’s not going on a button the same copyright and same biggest UGC short-form music-content music, sharing it, and upvoting the stuff they to happen overnight, just like past musical rights as to a professional musician? If platforms embedding their technology,” like best to help it reach a wider audience. revolutions have not happened overnight,” you look at the history of copyright, it was is how Cliff Fluet describes the potential It would be an ecosystem sitting apart from he says. invented in a world of scarcity, where only a impact of these acquisitions. the traditional music industry, although as “But it’s hard for me, from where I’m handful of people could access the means of HumTap is one of the first consumer- Phillips suggests, it might be a pool which sitting, to imagine a world where there isn’t production. A world where works and artists focused AI-music apps to explore all this. rightsholders can dip into, to identify the some combination of interactive technology, and authors were scarce,” says Reed Smith’s Besides turning people’s humming and creators who may not need the ‘training and lower or no-skill people become Sophie Goossens. finger-tapping into melodies and beats, it wheels’ for long. musicians and start creating content. And “In a world where making music can includes a feature to shoot a quick video It would certainly be disruptive. “When that’s going to have major implications happen at the push of a button, the volume that can be shared alongside the final tune. there are rules and laws in place that protect for the industry… We have never had an of songs created might challenge other For now, it’s the closest thing we have to scarcity, and you have no more scarcity, industrialisation of music creation before.” copyright concepts, including the concept the ‘TikTok for music-making’ app that’s that breaks it,” says Amadeus Code’s Taishi This is the kind of talk that won’t sit of ‘originality’ which is indispensable for being mooted. Fukuyama. “There’s a lot of big ideas around well with some musicians: it could be seen copyright protection to exist, at least as “They want to do for music what that, and I think that’s what fundamentally as positioning AI as a threat not because far as European copyright is concerned. Instagram has done for photography: the gets investors excited: when you propose an machines will replace musicians, but rather How copyright is going to be influenced by market might be teenage kids who want what’s happening with AI and AI-generated easy access to music composition, and music is a fascinating thing.” the ability to share it and have fun with it. Mitchell offers a more optimistic view. There are more people looking at how you “It’s very hard to imagine the ceiling on this make this more playful,” says Abbey Road’s opportunity once you really crack this: a Isabel Garvey, who got a close-up look at way of bringing more people into music- HumTap’s technology when it took part in making. My goal is not to replace artists: the studios’ startup incubator. that’s ridiculous! We’re going to make the “We often talk about this in the very overall size of the music industry larger,” serious sense: how to help professional he says. musicians. But actually AI also has the “If we can bring tens or hundreds of ability to unlock creativity for people millions of people into our world, it will without needing formal training on an create a bigger world, and it will foster instrument. That could be an educational innovation… And remember, there’s no

report piece, or a play piece.” such thing as AI music. It’s about people She goes on to note that the challenge making music. If you start with that lens, – as discussed earlier in this report – may everything makes a lot of sense.” :) the 16

ISSUE 425 25.11.19 AI REPORT Talking Point 8: AI gets embodied

he arguments over whether studio Betaworks, which made to live in – inviting fans in too. an AI could ever be a ‘bet- Auxuman one of the first cohort in “We think the next social media is like ter’ musician than a human its Synthetic Camp program earlier a game world,” he says. “We’re starting is tired, not to mention built this year. by becoming very good at building these onT a false premise that music’s “We see this as creating a characters, but the future might be these quality is objective, rather than pipeline of digital talent. We have large-scale simulated worlds. And then subjective. the means to build a collective these people are going to be there, and There’s another strand to that of engines and systems into one, you can visit them from time to time, and debate: that generative AIs lack in order to create digital talent,” make friends.” the ‘backstory’ of human artists – explains Koosha, who even before “Maybe that’s the Netflix of the future! everything about Adele or Beyoncé Auxuman had been experimenting Although yes, it’s still far-fetched for now, or Lewis Capaldi or Billie Eilish or with automating elements of his so we are working only on great digital [insert your own examples here] that creative and production process. talent. As people start to understand their *isn’t* their music, which forges their That’s now being used to create influence and communicate with them, we connection with millions of fans. the music for Auxuman’s five will start to build the case for that world.” One answer to this ‘faceless AIs artists, along with lyric-generating The fact that Koosha is an artist can’t be pop stars’ point is, well, algorithms driven by “internet life gives him a good perspective on the to put a face to them. We already – words that humans have said debates around creative AI and how it have avatar artists, most famously on the internet”. Auxuman’s first sits alongside human musicians. Indeed, Hatsune Miku from Japan, who compilation album was released he’s decided that his own next album will plays holographic stadium concerts earlier this year, with plans to have “no machine involvement” at all, in her homeland. Japan and regularly put out more music. scratching a different creative itch to his South Korea’s history with avatar singers “Yes, we fall into the category of ‘AI daytime role working with Auxuman’s stretches back to the 1990s. It’s wonky and wacky generative content’ like Endel and many digital-talent roster. More recently, virtual influencers like sometimes, but I hope it makes others, but we don’t make functional He’s clear about the key challenge for Miquela have made headlines with their you feel something music,” stresses Koosha. “We focus on the latter. “Music isn’t about academia. Instagram popularity – and in her case, experiential content. And we’re focusing Music is entertainment, a cool thing. You – Ash Koosha, Auxuman have also released tracks and music on Generation Alpha, the next generation, have to enjoy it, memorise it, on many videos. Meanwhile, games firm League who are just growing up, and who are going levels. That’s what I hope people feel about of Legends has created a series of virtual ‘AI-generated artists’: Yona, Mony, Gemini, to be more comfortable in being introduced Yona’s music,” he says. groups, including the recently-debuted Hexe and Zoya. to characters who are not human.” “It’s wonky and wacky sometimes, but hip-hop band True Damage. The company was co-founded by Another key point is that Auxuman I hope it makes you feel something. The Behind these avatars are humans musician Ash Koosha, designer and sees its characters as the company’s main first time you listen it may sound odd, but making the actual music, but what computer-generated (CG) artist Isabella product, rather than just the music. In the the second or third time, you feel like you

report if the next stage is adding AI to the Winthrop, and AI researcher Negar future, it has ambitions to enable fans to know her, and it’s a person. That’s the equation? That’s what London-based Shaghaghi, and is backed by $200k of interact with them one-on-one, and even path I’m trying to follow, and see where startup Auxuman is doing with its five funding from US venture-fund and startup to build a virtual world for the characters this goes.” :) the 17

ISSUE 425 25.11.19 AI REPORT Startup profiles Who’s who in the world of AI music? Here are some of the startups to watch

Accusonus In some ways, Boomy followed the pattern www.accusonus.com of early AI-music startups like Jukedeck This Greek startup recently raised a $3.3m and Amper, with a web-based tool to Series A funding round, five years after generate music using its AI. However, it launching. It makes tools for musicians has taken a step further: you can sign up, and producers, including its latest product create a bunch of tracks, compile them Rhythmiq, a ‘virtual instrument’ that into albums and then have it release them includes an ‘AI Beat Assistant’ designed to (under your name) on Spotify, Apple Music be used by artists, producers and DJs. and other streaming services.

AI Music Dadabots www.aimusic.co.uk www.dadabots.com A British startup that first came to Less a startup, more a “cross between notice with its plans for ‘shapeshifting’ a band, a hackathon team, and an technology capable of transforming ephemeral research lab” in their own tracks into a variety of genres and moods. words. Among the projects so far: a 24- Its experimental app Ossia launched in February 2019, but more recently the company has moved into ‘sympaphonic’ advertising: AI-powered ads that sound similar to each listener’s preferred music. Amadeus Code

Amadeus Code When it launched, US firm Amper Music through a set of five virtual www.amadeuscode.com was a rival to Jukedeck: you could sign characters. Yona, Mony, Another AI-music company with the up and use its AI to create music for Gemini, Hexe and Zoya flexibility to explore different models. Its your YouTube videos, games and other have been given their own first app was aimed at songwriters: an projects. Then the company shut down personalities (complete with AI-driven compositional tool to help them its beta composer to focus on ‘enterprise’ literary influences) and – escape writer’s block or try new ideas. products, including a partnership with backed by $200k of seed Recently, the company also launched a Reuters, and another with Tencent Music. funding from Betaworks – will production-music-library, Evoke Music, be releasing new music on a with tracks created by its AI. Auxuman monthly basis.

report www.auxuman.space hour stream of AI-generated technical Amper Music London-based startup Auxuman isn’t just Boomy death metal on YouTube; a similarly www.ampermusic.com creating music using AI: it’s releasing it www.boomy.com round-the-clock Twitch channel of piano the 18

ISSUE 425 25.11.19 AI REPORT Ones that got away Two AI-music startups have already music with startup Melodrive; been acquired by bigger fish Sheppard, LifeScore came out of stealth in and an AI-generated jazz Popgun February 2019. The company has recorded stream from space! Why? We’ve listed 18 startups in this section its own stems (in Abbey Road Studios) “Mischief”. of the report, but two more were taken which its app then stitches together out of the market by acquisition this year. into endless, adaptive music – including Ecrett Music Jukedeck kicked off the modern era of AI responding to the user’s movements and www.ecrettmusic.com music with its tool for generating origi- other inputs. Another startup from the nal music for video producers and other original Jukedeck / Amper creators. Melodrive Music template, armed with It was bought by TikTok’s parent com- www.melodrive.com the slogan ‘Royalty Free Music pany Bytedance this year, and its CEO Ed Melodrive started as a startup with an AI- for Creators’. Aimed at video- Newton-Rex is now working as director of music engine that could be used by games makers, games developers Bytedance’s AI Lab, although the com- developers to create original, adaptive and podcasters, it offers an pany’s plans for AI music (and specifically soundtracks for their games. Now, online tool to generate music whether they extend to TikTok) have yet it’s pitching itself as a firm offering ‘AI based on various contexts to be revealed publicly. consulting for audio-centric startups’, with and moods: from suspense, Another startup, MXX (originally plans to work for a few clients every year calm and chill through to Mashtraxx) was bought by another social- with coaching and workshops, through to ‘roadtrip’. music app, Triller, this autumn. MXX’s building prototypes and full systems. technology was less about generating Endel music, and more about using AI to edit Mubert www.endel.io tracks to fit videos – including for fans. www.mubert.com This startup made headlines Again, how its tech will be integrated Mubert’s AI is capable of generating earlier in the year with reports into Triller, which has 13 million monthly original music in more than 30 moods and of a ‘label deal’ with WMG, active users sharing lip-sync and dancing genres. Its business model is offering this although a correction was videos, remains to be seen. music via an API and streaming licence to quickly issued stressing that businesses: from fitness apps and mobile it was simply a distribution mobile AI-music apps. Humtap starts by games developers to social networks and agreement. Endel’s core getting you to hum a melody into your retailers / hotels requiring background business is an app (and smart-speaker smartphone’s microphone, which it then music. It also has a consumer-focused skill) generating music for relaxing, September 2019. It adapts music in real- converts into various instruments. You can mood-music app available. focusing and sleeping. However, the time according to signals including the then tap out a beat, film a video and share company has been releasing some of its time of day, the local weather, and the the results on social networks, with the AI Popgun music as albums too. movement of users. Fluent is keen for ensuring everything sounds polished and www.popgun.ai artists to upload stems to work with its professional. Australian startup Popgun started off with Fluent technology, too. an AI called Alice that could respond to www.fluentmusic.io LifeScore (and accompany) human pianists. Since

report One of the most recent companies in HumTap www.lifescoremusic.com then the company has taught its AI to this list. Fluent is a Russian startup that www.humtap.com Co-founded by Siri co-creator Tom compose and play bass, drums and piano; launched its ‘immersive music’ app in One of the first consumer-focused Gruber and composer and musician Philip it’s working on vocals; and has an Ableton the 19

ISSUE 425 25.11.19 AI REPORT

Vochlea Vochlea WaveAI www.vochlea.co.uk www.withalysia.com This British startup is behind Dubler, WaveAI’s Alysia app is pitched as both software that turns a human’s vocalisations ‘lyrics assistant’ and ‘melody partner’, (from beatboxing to “a hmmm, laaaa or creating tunes AND words for songwriters oohhh”) into music, with its Studio Kit – in similar territory to Amadeus Code. including the software and a microphone WaveAI has also been working on vocals: to use with it. The AI is what’s translating its AI can sing the lyrics it has created. the vocalisations into MIDI, working with Some of its recordings have been released various digital audio workstations (DAW) to commercially, including a Christmas suit musicians’ needs. album last year. The big guns Far from startups, some large tech companies are active in AI music

Google Magenta is Google’s open source project developing AI for generating songs, as well as images and other content. Its Magenta Studio plugins and NSynth ‘neural synthesizer’ have been pored over by startups and researchers alike. Google also launched an AI-powered ‘doodle’ celebrating Johann Sebastian Bach in March 2019, encouraging musicians to play with it.

OpenAI OpenAI is the non-profit AI company backed by Tesla’s Elon Musk and LinkedIn’s Reid Hoffman among others, exploring ‘general artificial intelligence’. But also music: in April 2019 it launched a prototype of Musenet, a deep neural network capable of generating songs with 10 different instruments, combining styles “from country to Mozart to the Beatles”, seeking feedback.

Live plugin called Splash Pro that helps video-creation industries. The pitch is that Facebook musicians generate AI music to work with this music can be “sculpted to The social network has poked at the idea of AI music. In May 2018, it revealed that on their own projects. perfectly match the on-screen action”, its AI Research department had worked on a “universal music translation network” with the startup building its own library capable of converting one type of music to another. “For example, we convert the Scored of tracks in a variety of genres and audio of a Mozart symphony performed by an orchestra to an audio in the style of www.scored.film moods. Its Scored Sync plugin will be a pianist playing Beethoven.

report Scored’s focus is on ‘dynamic music’, available for various pro video-editing created for clients in the film, TV and tools. the 20 Music Ally is a music business information and strategy company. We focus on the change taking place in the Music Ally is an ISSUE 425 industry and provide information and insight into every aspect of the business, consumer research analysing 25.11.19 example of perceptive the changing behaviour and trends in the industry, consultancy services to companies ranging from blue journalism at its chip retailers and telecoms companies to start-ups; and training around methods to digitally market your best, with unrivalled artists and maximise the effectiveness of digital campaigns. We also work with a number of high profile music Andrew Fisher, events around the world, from Bogota to Berlin and Brighton, bringing the industry together to have a good CEO, Shazam coverage of the commonsense debate and get some consensus on how to move forward. Entertainment digital music sector”

ANTHONY CHURCHMAN STUART DREDGE Contact: music:)ally Business development Editor at large Studio 11, Holborn Studios, 49-50 Eagle Wharf Rd, [email protected] [email protected] London N1 7ED Tel: +44 (0)20 7250 3637 b www.MusicAlly.com e [email protected] EAMONN FORDE ISABELLE LJUNGQVIST

Reports editor Digital marketing coordinator Registered company number: 04525243 [email protected] [email protected] VAT number: 858212321

© Music Ally Ltd. For the purposes of personal, private use the subscriber may print this publication or move it to a storage medium; however, this publication is intended for subscribers only and as such may PATRICK ROSS not be redistributed without permission. Subscribers agree to terms and conditions set up on the Music Ally website, except where a separate contract takes precedence. Music Ally has taken all reasonable Director, digital strategy endeavours to ensure the validity of all items reported within this document. We do not assume, and [email protected] hereby disclaim, any liability for loss or damage caused by errors or omissions. In particular the content is not intended to be relied upon in making (or refraining from making) investments or other decisions. We cannot be held responsible for the contents of any linked sites.

Our clients include: report the