Google has unveiled a new AI system called “MusicLM” that can create high-fidelity music in any genre just with a text description, according to a research paper. However, the company is fearing the risks involved and says that it has no immediate plans to release it.
Google is not the first one to try generative AI for songs. There have been several other attempts, including Riffusion (an AI that composes music by visualizing it), Dance Diffusion, OpenAI’s Jukebox, and Google’s own AudioML.
However, due to limited training data and technical constraints, none of them have been able to generate songs that are particularly complex in composition or high fidelity. MusicLM is expected to be the first that can.
Read More: Oreo Unveils Its Metaverse, OREOverse
MusicLM was trained on a dataset of about 280000 hours of music to learn to generate coherent songs based on descriptions of significant complexity, such as Berlin ’90s techno with a low bass and strong kick or enchanting jazz songs with a memorable saxophone and a solo singer.
Its capabilities extend beyond creating short clips of songs. Google researchers showed that the system could build on existing melodies, whether played on an instrument, hummed, sung, or whistled.