Google’s impressive new AI system can generate music in any genre given a text description, but the company is wary of its risks and has no immediate plans to release it. The AI system was revealed at the Google I/O conference last week, and is still in development. However, if it proves successful, it could revolutionize the music industry by providing listeners with an unlimited supply of unique tracks for free.
In recent years, there has been a renewed interest in the synthesis of music. Pioneered by classical composers such as Bach and Mozart in the 1700s, it was only recently that scientists and engineers started to explore ways to create original compositions using computer algorithms. Today, there are several generative AI systems available that can produce quasi-original pieces of music – but none have reached the level of sophistication or complexity achieved by traditional composers. Google’s MusicML system is perhaps the most well-known example of this type of algorithm, but it is hardly alone; other systems such as Riffusion, Dance Diffusion, and OpenAI’s Jukebox also exist. While these algorithms are able to produce interesting and sometimes complex pieces of music, they still lack the finesse and polish found in productions from more traditional composers.
MusicML is perhaps the first open-source music annotation platform that can create fragments of music in an understandable language. This makes it possible for people to share their annotations and track changes over time
Despite containing no labeled data, MusicML accurately generates coherent songs for descriptions of “significant complexity.” The algorithm seems to be factorizing the elements of each song into smaller and smaller pieces until it can create a rudimentary melody that matches the desired description. For example, one MusicML song describes an enchanting jazz song with a memorable saxophone solo and a solo singer. The saxophone solo is harmonically complex, while the singer has a low voice that contributes to the ethereal feel of the music. In contrast, anotherMusicML song describes Berlin ’90s techno with a low bass and strong kick. The bassline is thumping and provides an aggressive edge to the techno beat, while the kick drum accents various beats in 16th notes for added energy. These two songs are not without differences – one is lively and upbeat while the other is more driving – but they both contain identifiable elements that make them sound like human compositions.
The ability to generate realistic sounding music without a musician or instrumentalist in the loop is impressive, even if it doesn’t always nail down exactly what’s being described. Even when fed long and meandering descriptions, MusicML manages to capture nuances like instrumental riffs, melodies and moods. This could be an interesting alternative for creating soundtrack-like sounds without having to rely on an actual musician or any human input.
I was definitely lost in space listening to this song! It induces the experience of being lost in space, which is a great way to relax and wind down. If you’re looking for an ambient track that will put you in a good mood, I highly recommend checking out this song.
It’s a familiar sound to anyone who has played an arcade game- the high-pitched chirping of distant birds. The soundtrack is one of the main attractions for many players, and it creates that nostalgic feel that makes going to an arcade all the more irresistible. Some earworms can be traced back to this atmospheric music, and people often cite it as a contributory factor in their love of retro gaming.
MusicLM has the ability to create short clips of songs that are both creative and engaging. It can also generate narratives or story lines for videos, making them even more interesting and compulsive to watch.
Electronic music fills the air as gamers race around a virtual world, scoring points and grabbing treasure. Meanwhile, a healing meditation song floats by on the river beside them, lending an calming presence to an otherwise hectic experience. Suddenly explosives light up the sky, adding an unexpected but delightful touch of excitement to what had been beginning to feel like a routine day.
Despite its rough edges, MusicLM is a powerful tool that can generate impressive sounding music. With enough dedication and experimentation, any amateur musician could conceivably produce quality tracks using this program. It’s not perfect by any means, but it’s worth giving MusicLM a try if you’re looking for something to supplement your existing music collection or to create some original tunes of your own.
Despite the many ethical challenges posed by a system like MusicML, there are numerous possibilities for how it could be implemented, and its potential benefits are undeniable. Whether used to author new music or simply generate remixes of preexisting tracks, MusicML has the potential to create infinitely varied genres and styles at an unprecedented pace. While there are still many hurdles to overcome before this technology finds widespread use, its prospects look bright.
There is a risk of potential misappropriation of creative content associated to the use case, but the paper’s authors emphasize the importance of more future work in tackling these risks. In particular, they note that understanding how listeners interact with music generated through this technology will be crucial in ensuring safe use.
It is already illegal to rip or redistribute copyrighted music, so it is likely that MusicML will be seen as infringing on copyrighted material. This could create a huge legal issue for those who use MusicML, as videos of songs being created automatically using the system could be taken down. It remains to be seen how exacting the regulations will be around MusicML – but until this technology develops further, users may want to take care when uploading their creations online.
Sunray’s whitepaper argues that because AI music generators like MusicML create “tapestries of coherent audio from the works they ingest in training,” they are violating the United States Copyright Act’s reproduction right. This effectively means that, even if an individual user only uses MusicML to generate a small amount of music based on copyrighted material, they could still be infringing on intellectual property rights without knowing it. This is particularly troubling given the snowballing use of AI in all sorts of fields – from image- and code-generating services to music generators – and the lack of clear legal protections for creators when their data is used in this way.
Andy Baio’s musings on the practicalities of generating music using artificial intelligence raise some interesting legal questions. Although generated music may be considered a derivative work, it could be protected under copyright if used commercially. Fair use might also apply in certain cases, but case-by-case judgments would likely have to be made.
This litigation is just one example of how artists are currently fighting for their rights when it comes to music-generating AI. In the future, we may see more clear documentation of who owns what rights when it comes to these systems, which would be a positive development for both artists and software developers.