It seems that every day we witness new advances in artificial intelligences capable of generating content. After examples like DALL-E, which can create images from natural language descriptions, Google researchers have now managed to generate multi-minute musical pieces from this type of text description.
MusicLM can create music from natural text instructions, including prompts like “suitable for a prison break movie.”
The tool that made this possible is called MusicLM, and it can follow text descriptions regarding style, instruments, rhythm, and more—generating musical pieces that precisely match those instructions.
The results could easily pass for compositions by human creators, but Google’s researchers have also generated clips focused on specific instruments, from maracas to cello, or even musical pieces that could serve as soundtracks for movies or TV series, including prompts for “music suitable for a prison break.”
MusicLM can even generate audio clips that take into account the performer’s skill level, so the same piece will sound different if you ask for it to be played by a beginner or an advanced musician. In addition to musical instruments, it can also generate voices, though in this case the quality is more debatable.
For now, Google has no plans to offer MusicLM to the public, citing possible risks of what they call “potential misappropriation of creative content”, as well as other inappropriate uses like cultural appropriation. Instead, they have announced they will release a collection of over 5,000 examples of music created from text instructions, to serve as a reference for other musicians and technicians developing similar AI systems.
