Jessica Powell (Co-Founder/CEO, Audioshake) on music startup challenges and the "new possibilities" with music AI
Last week, we were joined for a Water Cooler interview with Jessica Powell, CEO of AudioShake.
Jessica is an author (The Big Disruption: A Totally Fictional But Essentially True Silicon Valley Story) and former Vice President of Communications for Google who cofounded AudioShake — a music startup that uses AI to break down songs into separate stems.
The idea came to Jessica, a lifelong musician, after a karaoke session with her cofounder. They’ve since raised several million dollars in funding, most recently led by celebrity investors like Metallica and Miley Cyrus.
We had a fascinating discussion which spanned the trials and tribulations of running a music startup, how AI can reshape how music is experienced, and how the traditional music industry can effectively forge partnerships with AI.
You can listen to the full audio below, or read on for key quotes and takeaways:
Music startups have a stubborn content problem
“The music industry is tough. If you’re trying to launch something in music, the general path is that you infringe someone else’s content, you grow off the back of content you’re not paying for, then you get sued. The lawsuit leads to a payout — which, by the way, never seems goes back to the artists — but also leads to a license, which normalizes the whole process.
Unfortunately it’s extremely difficult to launch something in the consumer space without being fully licensed. But who’s actually going to talk to you at the beginning, because you’re just a startup and you’re statistically likely to fail?“
Source separation is a musical application of AI that, notably, sidesteps licensing red tape
“AudioShake’s AI trains on thousands and thousands of real audio stems that teach our AI how to identify different instruments from a full mix — whether its mono or stereo — and separate those out into stems.
The thing is, with source separation, you don’t need to train on Elvis to be able to recognize and separate Elvis. We can easily train our models on library music, which tends to be transparent and easy to license by design. If we were approaching major labels and asking to train on their stems, I think that would be pretty difficult!”
AI won’t just allow artists to create more music — it’ll also allow their music to appear in more places
“Sync is a huge growth area for audio stem separation. It’s pretty cool to hear someone say that you opened their music up to new possibilities. The music in the trailer for Encanto, for example, is an AudioShake instrumental of an iconic Colombian song that they originally only had in an MP3 of the full mix.
It also opens music up to immersive mixing — in spatial audio, for example, you need to separate out each element of a song so you can place the sound objects in different perceptual fields. I love the idea that music will gradually become more immersive, and that fans will be able to interact with music in very subtle but much deeper ways.”
Expect to see more “non-controversial” use cases for stem separation in the future
“I think we’re going to see lots of use cases for stem separation where there’s revenue opportunities for artists and labels, and where the integration isn’t super controversial. For example, why shouldn’t karaoke exist in the original recordings and have artists make money from them? That’s an area where a label might be quicker to license stems. Another example is interactive music, allowing music in gaming platforms to change subtly as the player moves. But I think we’ll see all of these use cases form before we’ll see a fully fledged stem marketplace. I’m pretty confident about that.”