All these years when the alarm bells rang that robots were coming for our jobs, the creative types stayed cool. Sure, machines could count faster, stack better, drive a car. But they couldn’t compose a poem, put it to music, sing.
Now they can. Microphone drop.
To be fair, AI programs can no more create music than they can paint, choreograph, or trade actions without human intervention. But as humans train machines to study more complex patterns, words and sounds, the resulting AI-based software is starting to produce surprisingly creative results. Find out what’s out there and probably already on your playlists.
Lyric Jam: They’re just words…and now a bot can write them for you. At Canada’s University of Waterloo, researcher Olga Vechtomova, working with graduate students Gaurav Sahu and Dhruv Kumar, last year developed a system capable of generating lyrics for live instrumental music. It basically analyzes the genre it’s hearing, matches it to other songs it’s heard in the same style, and offers, in real time, a mix of phrases depending on the mood.
{{^userSubrated}} {{/userSubrated}}
“The purpose of the system is not to write a song for the artist,” Vechtomova said in a press release. “Instead, we want to help artists realize their own creativity. The system generates poetic lines with new metaphors and expressions, potentially leading artists in creative directions they haven’t explored before.
Try it on Lyricjam.ai. I hummed a half-remembered hymn instead of recording an original tune. The machine suggested moody phrases involving dream, silence, rain, and moon. Not the best, but definitely less annoying than Baby Shark.
Boom: Alex Mitchell’s website and app helps everyday people, even the deaf, create and edit their own music. Join Boomy.com for free. Choose a genre (Electronic Dance, Lo Fi, Rap Beats and others) and filter the styles inside each. The AI does all the hard work, composing a song in around 30 seconds.
{{^userSubrated}} {{/userSubrated}}
Users can add vocals, change instrumentation, tempo and arrangement. Submit a recorded song to a streaming platform or, say, a video editing app, and you could earn royalties, although Boomy retains the rights to the music.
Boomy went live in 2018 but found its moment of glory in the long empty hours of the pandemic, when users recorded the majority of Boomy’s 3 million songs. My song, TheGreaterBombay, draws from his Global Groove selection and contains Latin filters. It sounds unmistakably computer-generated, but also sounds eerie, much like a theme music for a Narcos spin-off.
OpenAI jukebox: Humans are rarely original, so why should AI be? Jukebox accepts genre, artist and lyrics as input and generates a new music sample. At best, the songs sound like fillers from Elvis Presley’s B-side or Katy Perry’s catalog, vocals, guitars and all. At worst, they look like something an underpaid DJ might rush out. The neural network can also work with a sample or snippet and complete a song. Check out examples at openai.com/blog/jukebox.
{{#elements}}
{{/elements}}
Dadabot: On its YouTube channel, this program is described as an “AI death metal band. Emulate dead or alive musicians with neural networks”. The fake band plays real music 24/7. The humans behind the AI are metalheads CJ Carr and Zack Zukowski, who studied machine learning at Berklee College of Music. They use material from existing death metal, mathcore, and skate punk bands to power the AI-generated approximations of Dadabots Online.
“While we set out to achieve a realistic recreation of the original data, we were delighted by the aesthetic merit of its imperfections,” they write, in a 2017 article published on arXiv that breaks down their process. “Solo singers become a lush chorus of ghostly voices, rock bands become crisp cubist jazz, and crossovers of multiple recordings become a surreal sonic chimera.” The band’s website, Dadabots.com, lists nine AI music albums curated by the two humans.
{{^userSubrated}} {{/userSubrated}}
Authentic artists: Virtual bands have been a part of mainstream music since at least 2001, when Gorillaz — a four-member troupe represented by animations, but with a lone composer behind it all — released their first album. But Authentic Artists’ 12 AI-powered virtual musicians (they range from a lo-fi-loving cyborg to a high-octane, half-iguana DJ) even compose their own music.
Site users offer information on what they want their virtual concerts to be: an artist, a genre, a movie soundtrack or even another concert. Avatars then generate compositions and adjust tempo, volume, and even skip to the next song based on audience feedback, almost like a gamified concert. Three virtual artists (essentially programs), Nayomi, DJ Dragoon and Gnar Heart, also now perform as the “collective” WarpSound. Why? To “co-create real-time live experiences with audiences around the metaverse,” explains the company’s website, authentication-artists.ai.
{{^userSubrated}} {{/userSubrated}}
How bad is your Spotify: It happens quite often with humans, so it was only a matter of time before musical robots turned into music snobs. While year-end reviews like Spotify Wrapped are usually upbeat, telling you which songs and genres you’ve been listening to the most, a bot will judge your musical tastes throughout the year.
Log in with your Spotify or Apple Music account. Let the AI draw on its roughly 76,000 parameters of what humans objectively consider good and bad music (reviews, store recommendations, etc.) to examine your playlists. Then the ribs begin. Did you really stream Led Zeppelin without irony 79 times this month? You’ve been listening to Tracy Chapman a lot, are you okay?
For the final judgment: machine-learned multi-hyphenated text that’s perfect for your reading list. Was my 20th century reading list bad; hot topic-vomited-on-you bad; or probably-voted-for-a-bad-fictional-character? I do not say. Access your own insults at Pudding.cool/2021/10/judge-my-music.
{{^userSubrated}} {{/userSubrated}}