Google is a major player in AI. It may be that ChatGPT has Google a bit frazzled by its package of services, or not. Whatever the case, Google has suddenly become very visible for its MusicLM AI, which can create music in a range of styles from text prompts. The prompts are called “rich captions”.
The really big surprise is that this isn’t as bad as it looks. If anything, MusicLM is more hampered by the quality of its sample sounds and the lazy sort of recording mixes that infest current pop music.
The music that MusicLM makes is OK in the same sense a beginner trying to mix recorded sounds is OK. It’s a bit clumsy, and the pitch of some notes is pure 1990s MIDI a bit too often. The most obvious flaw is that these sounds need proper pitch.
With these serious limitations, the sound made by MusicLM is forgivable, and at roughly the same standard of its inputs. That’s not great, but obviously can be improved pretty easily.
At fundamental sound generation levels, a musical note is given specific values, then used to generate sound using an oscillator. That’s why MIDI sound is so often simply not up to scratch. Modern sound is closer to true pitch and can be heard on professional composer software like Hollywood Strings.
The music box of the future, or what?
As with a lot of AI “breakthroughs”, this involves hooking up what is basically a chatbot to other functions. Having muttered which – MusicLM is flexible, and credit should be given for what is obviously a vast amount of back-end work on making the text link work to deliver new music generation.
This process is literally ditch-digging with oscillators and GHz. There are so many sounds to be duplicated and put into context to create a piece of music. Sure, you can link to patches, etc., but you’re always going to get hit with sound quality issues and non-sequitur noises in styles, even with something as basic as a single drum beat.
So on that basis, with so many handicaps, MusicLM is at least a good basic platform. It’s not the AI’s fault that the world is awash with crappy patches. The big issue here is how useful is MusicLM as a musical creator and musical asset to musicians?
Qualifier – I’ve been playing music for a while, and occasionally threaten to make musical noises. I know all of the original synthetic sounds and how they evolved as workable recording assets pretty well, too. This op-ed isn’t a rush to judgment for precisely that reason.
Therefore, some musical heresy is forthcoming. MusicLM could well be extremely useful for musicians on many very basic levels:
- Managing your own musical doodling. MusicLM could easily be geared up to write music, too. In effect, you’d be collaborating on musical text. (Could save a lot of work on musical annotation, too.
- Jamming with AI. Jamming is extremely good practice. Your fluency has to improve. Sharpen up your skills on whatever you like, and have fun. Your text description could be “orchestral metal” or “1930s heavy brass” or “the reggae version of Sound of Music”.
- Basic recording. The AI does the lumbering, time-consuming stuff and you do your stuff as overlaid tracks. MusicLM is quite capable of delivering on this aspect. Also removes the sheer tedium of repetition. That’s pretty much how a lot of modern music is made, but MusicLM seems well able to deliver more range and scope than idiotic “categorized” patches, etc.
- Filler? What filler? The old word “muzak” isn’t appropriate here. This AI seems to have at least the capacity to produce enough scope to create an aesthetic, with enough chops (style) to bring it off. Simply because it’s an AI, it doesn’t have to stick to any boring, predictable, formula. For recording purposes, those with enough patience should be able to make it work.
- Ambience, maybe. The most misused word in the English language is probably ambience. Because it’s such a vast term, the word is abused on a routine basis, particularly in music, where it’s basically meaningless without a lot of added verbiage. This thing, by sheer luck, perhaps, is based on terminology rather than babble. It will define its own ambience.
Copyright, ethics and pure drivel
It seems that one of the issues reported to be holding up the release of MusicLM on the same terms as ChatGPT is “ethics.” Google is worried about its impact on arguably the most unethical industry on Earth, the music industry.
This poor martyred industry, hilariously reported as “worth $35 billion” (presumably they left out the generations of production budget ripoffs and money laundering) might be impacted? The music industry has no interest whatsoever in music of any kind and has spent the last 60+ years proving it. They’ll be fine with it when they figure out how to make money out of it.
That’s the main reason so much crap is produced. These are transactions, according to the industry. Fun, happiness, insights, emotions, and the rest of the experience don’t fit on spreadsheets.
Then there’s copyright. This is the classic insane obese self-worshiping bastard of the creative arts. It’s almost a religion, and it breeds lawsuits. From what I’ve heard of MusicLM, and learned about how it makes music, there’s a legal problem with even the suggestion of any copyright infringement here:
- MusicLM isn’t a legal, corporate, or natural person. There’s nobody to sue.
- You can’t even prove intent to infringe on a copyright.
- You also can’t accuse anyone of “copyright infringement by text description of a piece of music”. (Partly because even an identical description would and could not generate identical pieces of music.)
- Anyone who thinks you can wade through what’s likely to be billions of terabytes of AI-generated music and find a lawsuit is far more than an optimist.
MusicLM is not the music box of the future. Ironically, it might just save music from people who should have nothing to do with music.
_____________________________________________________________
Disclaimer
The opinions expressed in this Op-Ed are those of the author. They do not purport to reflect the opinions or views of the Digital Journal or its members.
