Google demonstrated an AI system that generates music from text descriptions, but the search giant has no plans to release the AI due to potential risks. Google’s AI system is called MusicLM. Similar AI music systems include Riffusion, Dance Diffusion, Google’s own AudioML and OpenAI’s Jukebox. No AI system has previously been able to compose complex or high-fidelity songs due to technical limitations and limited training datasets. MusicLM may be the first to do so. Google researchers used a 280,000-hour music dataset for training, learning to generate coherent songs from complex textual descriptions. It creates songs that sound like something a human musician might create. The researchers also noted an ethical challenge for the MusicLM system: the tendency to incorporate copyrighted music from the training dataset into the generated music.
This article is transferred from: https://www.solidot.org/story?sid=73967
This site is only for collection, and the copyright belongs to the original author.