cj carr sonar

That AI should hog the limelight at Sónar +D, Barcelona festival Sónar’s “international congress” of music and technology, is no surprise.

That a broadly positive – and even lucrative – picture of the possibilities of AI should emerge from the event’s opening day is perhaps more unusual, including the tale of an emo band who earned a six-figure sum from the release of an AI-generated album.

This fascinating tale came from CJ Carr, head of audio research at Stability AI, the company behind popular text-to-image AI Stable Diffusion, but also music-focused project Harmonai. The latter develops open-source generative audio tools “to make music production more accessible and fun”.

Carr gave an early-afternoon Masterclass at Sónar on using AI technology to create music. In doing so, he drew on his experience with Dadabots, a 10-year-plus project that trains raw audio neural networks to generate music.

Examples he cited included using AI to create 10 hours of music trained on Swedish extreme metal band Meshuggah; a parody of skate punk band NOFX’s ‘Punk in Drublic’ called ‘Bot Prownies‘; and ‘Relentless Doppleganger‘, a live stream of AI-generated death metal that went viral.

YouTube video

Carr also talked about his use of OpenAI’s Jukebox, a neural net that generates music, including rudimentary singing. Among his experiments with this was an “impossible cover” of Frank Sinatra singing Britney Spears’ ‘Toxic’.

After being released on YouTube, the song was subject to a takedown notice, which Carr decided to contest, alongside the Electronic Frontier Foundation. The EFF, Carr explained, made the argument that this “cover” was fair use. The case was eventually settled in “YouTube court” between labels and YouTube with the copyright strike being reversed. (Although without setting a legal precedent.)

For those interested in the potential financial benefits of AI to musicians – a thorny subject in 2023 – Carr explained that Canadian emo band Silverstein had earned “six figures” from ‘Fake Feelings‘, a 1,000-song album made by Dadabots using AI trained on their music, which was then sold on the blockchain, song by song, to individual consumers.

“I think we are going to see a lot more of that,” Carr said. “We made six figures for that band.”

“In a sense we are entering a new era of sampling,” he concluded, explaining that OpenAI’s Jukebox allowed users to create songs that sample a specific era of music – eg 1970s funk – without copying a specific song.

Earlier in the day Pau Garcia, founder and CEO of Domestic Data Streamers, a Barcelona collective of journalists, researchers, coders, data scientists and designers, gave a talk focused on the creative and ethical implications of the use of AI in culture and the arts.

Garcia’s focus was on trying to escape from a polarised view of AI, whereby the technology will either will solve all our problems or kill us all. This polarised view, he said, is dangerous.

“There are alternatives – a critical perspective, that isn’t polarised – [which recognises that] there are dangers but potentially positive outcomes [from AI],” he said.

AI, Garcia said, can be problematic: whereas a search engine will give hundreds of thousands of answers to a specific question, AI will give one reply. But it is also a technology that, if explored correctly, has a great deal of potential in spaces that we might not have even considered.

Garcia gave the example of his company’s Synthetic Memories project, which uses AI technology “to build up new spaces for memories, to recreate spaces where visual memories have been lost” (according to the company’s website).

In Barcelona the company has been using AI with older citizens to recreate images of spaces and objects that they remember from their childhoods, of which no actual records exist, from a long-lost coffee pot, to a view from an old balcony.

YouTube video

Sónar +D was also home to a masterclass from the digital artist and founder of Rhizomatiks creative studio Daito Manabe, who, alongside engineer Yuya Hanai, explained some of the workings of generative AI and prompt-based image synthesis technology.

Using the examples of Rhizomatiks’ own Mitsua Diffusion CC0 text-to-image diffusion model l and the company’s “AI YouTuber” Mitsua Elan, Manabe and Hanai explored some of the creative implications of Rhizomatiks’ work. 

Sónar +D continues tomorrow (Friday June 16) and Saturday (June 17). On Friday, the programmes includes the final two panels from the AI & Web3 Creatives Summit, organised in collaboration with Barcelona’s Pompeu Fabra University, exploring expanded media and creativity under Web3, as well as a keynote from UK DJ and label owner Elijah.

On Saturday, the Sonar +D programme includes a talk from Weirdcore, a visual artist known for his collaborations with Aphex Twin and M.I.A., and a celebration of local talent, headed by journalist Aïda Camprubí.

EarPods and phone

Tools: platforms to help you reach new audiences

Tools :: Wyng

Through Music Ally’s internal marketing campaign tracking, we’ve recently discovered an interesting website by the…

Read all Tools >>