The AI arms race has a new soundtrack, and OpenAI wants to be the one writing it.
OpenAI is developing a generative music tool capable of composing music from text or audio prompts, according to reporting by The Information. Though the company has not yet formally announced the tool or confirmed its release strategy, sources say the initiative marks a significant pivot toward creative audio applications for the $500 billion tech giant.
The company is reportedly collaborating with students from New York’s prestigious Juilliard School for musicians, tasking them with annotating sheet music to help refine the model’s training data. While sparse on specifics, the project hints at a broader ambition to merge professional-level musicality with machine learning.
The endeavor places OpenAI in the crosshairs of a rapidly evolving and increasingly contentious space. Google has been testing the waters with its MusicLM model while Meta earlier this year released open-source tools for music generation under the Audiocraft moniker. But perhaps the most visible (and controversial) is Suno, a platform that allows users to create full songs from simple text prompts.
OpenAI itself is no stranger to generative music. The company in 2020 released Jukebox, a neural net capable of creating raw audio music complete with vocals. While experimental and ultimately discontinued, Jukebox foreshadowed the deeper AI-audio ambitions now taking shape under CEO Sam Altman.
Still, with great scale comes scrutiny. OpenAI has already been entangled in a web of copyright litigation, including lawsuits over alleged unauthorized use of copyrighted content for training language models. The company has inked some licensing deals, most notably with Shutterstock, which includes access to its music catalog, but the lack of clear details around how OpenAI is sourcing audio data for training its music tool is likely to raise familiar legal challenges.
The post OpenAI Is Developing a Generative Music Model: Report appeared first on EDM.
Recommended Comments