AI Can Break Songs Into ‘Musical Stems’ — Here’s Why It Matters
Artificial intelligence tools aren’t just generating new songs (and artists), they’re also plundering past hits for new revenue opportunities. Disney Music Group, for example, has agreed to let AI music startup AudioShake, a participant in the 2024 Disney Accelerator program, generate stems from his classic hits for use in new creative and marketing opportunities.
So don’t be surprised if you see old songs from Mickey Mouse and his friends in new remixes and applications in games, music, commercials and anywhere you are confronted with sounds.
Here’s everything you need to know about AI-generated stems, who’s using them, and what they could mean for the music industry and all music lovers.
Read more: Google unveils music AI sandbox that creates loops from prompts
What is a voice in music?
When talking about music, a voice refers to isolated song elements such as vocals, bass lines, drums, and other instruments. These parts of a melody are easily separated in digital music, where they are already built into separate tracks.
Traditionally, they were harder to capture with older footage taken before the invention of such technology, but AI has now completely changed that.
Read more: Making the Next Beatles: How AI Is Changing Pop Music
Vinyl DJs were the first
Digital stems have been available for DJs around the world to buy and use at home and in professional settings for almost a decade. But it’s only been in the last few years that you can use software to create your own stems. Having vocals, basslines, beats and other elements on their own opens up new dimensions of creative potential for mixing music.
Today, DJ software programs like Traktor and Serato allow any user — beginner or pro — to generate stems on the fly. Decades before this was possible digitally, and with the focus still on the analog era of vinyl, hip-hop, and dance music, DJs using vinyl produced and distributed records with instrumental and a cappella versions of their songs. Unauthorized bootlegs of old soul, disco, and house tracks with isolated vocals and instruments were easy to find in independent record stores.
That’s a big step up from what’s possible now, but the results of using digital software to create stems are mixed, especially when you apply such tools to older music that wasn’t created with digital equipment.
The magic is not done in every song using digital software to extract stems. Sometimes other elements of the song seep into the vocals or other parts and do not provide the drama, humor or other impact that a DJ can achieve with his own imagination and good technique.
The role of AI in stem creation
A growing number of companies like AudioShake, BandLab, and SongDonkey offer AI voice separation tools. These programs have algorithms that analyze the musical elements of a song and split them into different files.
AudioShake makes two AI-powered products: one uses AI to separate dialogue, music, and effects from movies and TV, and the other uses AI to separate instruments in songs into stems. It also generates lyrics.
“We were deeply impressed with AudioShake’s sound separation technology,” said David Abdo, CEO of Disney Music Group, in a July press release. “We are excited to expand our existing voice separation work and integrate AudioShake’s lyric transcription system. AudioShake is a great example of cutting-edge technology that can benefit our artists and songwriters.”
A voice-assisted musical future
Major recent corporate deals, such as Disney Music Group’s agreement to have stems made of its classic catalog, are a sign that companies with large music archives are investing in new ways to deal with this.
Prepare for an even more intense nostalgia experience than it already does in commercials, cinemas and even the supermarket. Expect the unexpected as AI becomes more widely deployed.
Perhaps the waves of interest in a particular era are no longer strictly chronological and we’re all being transported back to Mickey Mouse’s Fantasia.