Any type of music may be produced by Google AI with a written description

Forget ChatGPT, the future of AI content production may lie in music. Recent research on MusicLM, a system that generates music in any genre using a written description, was released by Google. The creation of AI music is not new. Projects like Google’s AudioML and OpenAI’s Jukebox have taken on the challenge, as Brozlex points out. However, the model used by MusicLM and its huge training library (280,000 hours of music) enable it to create music with a remarkable depth and variety. You could only enjoy the result.

Not only can the AI mix genres and instruments, but it can also create songs utilizing abstract ideas that are often hard for computers to understand. MusicLM can create a reggaeton-dance hybrid with a “spacey, ethereal” melody that inspires a “feeling of wonder and amazement.” Even the description of a painting or humming may be used by the technology to create tunes. A DJ set or soundtrack may be created by weaving together multiple descriptions in a narrative mode.

As is the case with many AI generators, MusicLM has issues. Vocals frequently sound odd in compositions and might be difficult to understand. The performances are better than you might imagine, but they may also be monotonous in ways that human efforts might not be. You shouldn’t anticipate an EDM-style drop or a traditional song’s verse-chorus-verse structure.

Simply don’t anticipate needing the technology anytime soon. The researchers aren’t making MusicLM available to the general public because to copyright issues, as they do with other Google AI generators. Approximately 1% of the music created at the time of release was directly plagiarized from the training tunes. While the licensing issues surrounding AI music are still up for debate, a 2021 whitepaper by Eric Sunray, who is currently employed by the Music Publishers Association, indicated that AI music may infringe on reproduction rights if there are enough “coherent” remnants of the original sounds. Similar to musicians that use samples, you could need approvals to distribute music made by AI.

In music, AI is already in use. Algorithms have been employed by artists like Holly Herndon and Arca to create albums and soundtracks for exhibits. But such either involve collaboration (as with Herndon) or purposeful irrationality (as with Arca’s). Although MusicLM may not be ready for prime time, it offers a glimpse into a period when AI may play a bigger part in the recording studio.