AI music generators could make the next hit

The Future.

AI-powered platforms are turning their generative capabilities to music, spitting out clips created from an algorithm trained on hundreds of hours of existing music. It’s murky water for the music industry, which is famous for its strict copyright laws. But forward-thinking musicians could use these tools as another instrument in their arsenal to craft sounds that we have yet to think of.

AI on FM
Stability AI is tuning its algorithm to craft new music.

• Techcrunch reports that Harmonai, backed by Stability AI, is beta testing a new tool called “Dance Diffusion,” which can generate “original” song clips.
• The tool creates these clips through the process of “diffusion” — generating new data by “learning how to destroy and recover many existing samples of data” (i.e., all the music from a specific artist).
• For now, the tool is only being trained on songs in the public domain that artists have contributed and music that is licensable under Creative Commons.

The tool is still pretty limited in that the inputs it takes in will spit out a clip similar in style. They’re also only a few seconds long, and the lyrics, if they have any, are mostly gibberish.

Soul of a song
Stability AI isn’t the only company experimenting with AI music creation.

• OpenAI, the company behind DALL-E 2, released “Jukebox” a few years ago, which can create whole songs when a genre, artist, and some lyrics are inputted… but the songs come out structureless and have “nonsense” lyrics.
• Additionally, Google announced AudioLM, which can extrapolate entire piano generations just by being given a “short snippet of playing.” So far, the tool hasn’t been open-sourced.

While musicians may bristle at the potential pitfalls of this tech, some are doing their best to harness it and even control it. A company called Spawning (run by technologist Mat Dryhurst and musician Holly Herndon) is working on a tool called Source+ that will allow musicians to opt-out of the databases these AI tools run on.

Meanwhile, Herndon released a tool called Holly+ that allows artists to include an AI version of her voice in their music — a glimpse of what the future may hold for AI-driven creativity.