It's not that there isn't enough electronic music being made, it's that every new track that lands on soundcloud is a drop in the ocean of mediocrity. There is _too_ much, and 99.9% is just boring to listen to, because it sounds like everything else. I listen to a LOT of electronic music (and have, since the mid 90's) and just don't have the patience anymore to sit through hours of average material to find one or two truly inspired artists.
I doubt I would turn to AI much for anything other than background noise while focusing on work. In fact, that sounds like a perfect use case for me. "Dear GPT, please compose a four-on-the-floor downtempo progressive track with soft pads, no vocals, and zero goddamned fake vinyl noise that runs for two hours straight..."
> It's not that there isn't enough electronic music being made, it's that every new track that lands on soundcloud is a drop in the ocean of mediocrity. There is _too_ much to listen to, and 99.9% is just boring to listen to, because it sounds like everything else.
Yep. this is why I don't feel like AI used in this manner moves the needle for music: people only actively listen to the best 0.1% of music anyway. The ability to create music that is firmly in the other 99.9%, as this stuff very clearly is, just means that the ocean of mediocrity has more water dumped into it.
> It's not that there isn't enough electronic music being made, it's that every new track that lands on soundcloud is a drop in the ocean of mediocrity. There is _too_ much, and 99.9% is just boring to listen to, because it sounds like everything else.
To the extent that sounding like everything else is a problem, how is ML generated music not going to have it?
And in general this isn't going to be a qualitative improvement in experience. ML algorithms for recommendation are searching the preference space in much the same way ML generation would, they're just doing it over existing stuff. If you really find 99.9% of existing material boring you're probably going to find a similar order of generated material boring.
Though I suspect 99.9% is hyperbole. My rate of "this is listenable and interesting and I'd like to come back " on Soundcloud is better than 1 in 25 on the worst day and better than 1 in a dozen on most, and the rate is often north of 1 in 6 for curated platforms like Pandora. It's never been easier to discover good new music to listen to with not much in the way of effort.
"To the extent that sounding like everything else is a problem, how is ML generated music not going to have it?"
AI generated art has explored all sorts of weird spaces that few humans have touched.
It's not difficult to make computers create unusual, original, bizarre work. The difficulty comes in making it both original and enjoyable/interesting.
Also consider that AI-generated music is often going to actually be a collaboration between a human and an AI. The human will be acting at least as a curator, because not everything created by AI is going to be pleasing, so some selection and catering to human taste will be required.
> The human will be acting at least as a curator, because not everything created by AI is going to be pleasing, so some selection and catering to human taste will be required.
Yes, and keep in mind humans are already doing this! It's very common to do tweaking of knobs on a synth/VST while recording and create a 10-20 minute audio file, commonly called a bass jam or mud pie, then select the best bits to use in a song. And of course, people use randomization tools to tweak the knobs for them. IMO use of AI to support this type of workflow is far more promising than going directly to the finished product.
I generated a 1:20 sample using your prompt "four-on-the-floor downtempo progressive track with soft pads, no vocals" using the audiocraft-webui fork, which allows for longer generation by overlapping generations.
It was not super discerning listeners (like it sounds like you are) that I meant to address in that second question. Sorry if that was not clear. Rather it had sounded from some of the comments that people were desperate for original tunes (and maybe not necessarily the most highly produced). But I didn't point to a specific comment, so maybe that's my fault.
We may also disagree on how much good stuff there is coming out, but I agree there is a lot of noise.
I doubt I would turn to AI much for anything other than background noise while focusing on work. In fact, that sounds like a perfect use case for me. "Dear GPT, please compose a four-on-the-floor downtempo progressive track with soft pads, no vocals, and zero goddamned fake vinyl noise that runs for two hours straight..."