Google has unveiled its latest AI breakthrough, MusicLM, capable of producing music in any genre with just a text description. Although not the first AI music generator, MusicLM stands out with its vast training database of 280,000 hours of music and advanced model, producing music with remarkable diversity and quality. According to TechCrunch, MusicLM’s output is worth a listen. MusicLM can not only blend genres and instruments, but also generate tracks from abstract concepts. The AI can create a hybrid of dance and reggaeton music with a “spacey, otherworldly” tune that elicits a “sense of wonder and awe.” It can even make melodies based on humming, whistling, or painting descriptions. The story mode allows for piecing together multiple descriptions for a DJ set or soundtrack.
MusicLM: google almost solved music generation
MusicLM Samples
https://google-research.github.io/seanet/musiclm/examples/
Despite its impressive abilities, MusicLM has limitations, like any AI generator. Some compositions may sound odd, and vocals can be difficult to understand. Although the performances are better than anticipated, they may also become repetitive. Do not expect an EDM drop or traditional verse-chorus structure in MusicLM’s creations.
What if it comes out MusicLM?
If MusicLM or similar technology becomes publicly available, it is likely to raise major legal concerns, even if marketed as a tool to aid artists rather than replace them. This has already happened with simpler AI systems. In 2020, Jay-Z’s record label took legal action against YouTube channel, Vocal Synthesis, for using AI to produce covers of Jay-Z’s songs, such as Billy Joel’s “We Didn’t Start the Fire.” YouTube initially removed the videos, but later reinstated them after determining the takedown requests were “incomplete”.
