Deep Learning Rocks How 2023’s soundtrack became AI-generated

Reading time
2 min read
Deep Learning Rocks: How 2023’s soundtrack became AI-generated

Fans of AI-driven music pressed play, while a major recording company reached for the stop button.

What happened: AI grabbed listeners by the ears when it helped produce a new single by The Beatles, mimicked the voices of beloved stars, and generated music from text prompts.

Driving the story: AI hasn’t quite had its first hit record, but developments in generated music put both fans and the record industry on notice that it may not be far away. 

  • Giles Martin, son of the producer of The Beatles’ classic 1960s records, used a proprietary audio demixing algorithm to pick apart a crude recording of an unreleased song by deceased band member John Lennon. Martin isolated Lennon’s voice so the surviving members could add fresh instruments and vocals. The result put the Beatles at the top of the UK music chart for the first time in more than 50 years.
  • Talented fans used voice cloning models to produce soundalike recordings in the styles of well-known artists such as Drake and Oasis.
  • Experimental pop star Grimes enabled the public to transform their own singing into a likeness of her voice, resulting in more than 300 faux-Grimes productions. Korean pop artist Midnatt used a similar system to translate a vocal track into five other languages.
  • In September, Stability AI released Stable Music, a diffusion model that generates up to 90 seconds of music or sound effects from text prompts, for paid public use. Stable Music followed Google’s MusicLM, a text-to-music model based on the transformer architecture.

Industry crackdown: Universal Music Group (UMG), which accounts for nearly one-third of the global music market, reacted swiftly to the wave of generated music. It blocked streaming services from distributing fan-made, voice-cloned productions and demanded that they block AI developers from downloading music by UMG artists so they can’t use it to train machine learning models. Shortly afterward, UMG partnered with Endel, a startup that generates background music. UMG artist James Blake released music he created using Endel’s system.

Where things stand: Generative AI is poised to play an increasing role in recorded music. AI-powered tools exist for many phases of recording production, including composition, arrangement, and mixing. The recent agreements between actors and writers and Hollywood studios may offer pointers to musicians and recording executives who would like to use these tools to make exciting, marketable music.


Subscribe to The Batch

Stay updated with weekly AI News and Insights delivered to your inbox