> a piano and cello duet playing a sad chambers music
No such thing as chambers music or a music. Maybe "a sad piano and cello duet" or "sad piano and cello chamber music" would be a better prompt.
MusicGen: Simple and controllable music generation
241–250 of 340 posts
Re: MusicGen: Simple and controllable music generation
#242(This is not meant to be an anti-ai-generated-art rant. It's coming whether we like it or not. But some of the motives in this thread confuse me.) Music producer here with an honest question to those saying "this will provide me with a simple soundtrack/background music for $PROJECT" Have any of you checked out / made offers on music production subreddits? Or other music subreddits? various music production discords?…
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…
> 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.
Re: MusicGen: Simple and controllable music generation
#243Surely cherrypicked, but holy cow. Where will this end? Can you imagine what the HN front page will spit out in a few years?
> Can you imagine what the HN front page will spit out in a few years?
A profitable startup? That might be asking too much though
Re: MusicGen: Simple and controllable music generation
#244Earlier quoted context omitted.
Like I said, cynical/jaded internet commenter - I don't need Mick Gordon's dynamic compositions, I just need some background music for my game that's good enough. And no you can't do this already: const musicSpeed = inFightSequence ? 'intense' : 'chill'; const musicPrompt = `drum and bass beat with ${musicSpeed} percussions`; playMusic(musicPrompt);
I wish you luck with your game which constantly oscillates between https://youtu.be/dLkFh9Kn8AU?t=60 and https://youtu.be/gQktj-WkgEo?t=75 because you didn't put any thought into the music beyond telling the computer to make "chill DnB" and "intense DnB".
I dunno man, if it output those tracks with a smooth transition and some volume matching i think most game developers would call that a success.
Re: MusicGen: Simple and controllable music generation
#245Earlier quoted context omitted.
When Animal Crossing blends between a different composition for every hour of the in-game clock, with variants for different weather conditions, is that not changing based on "time of day, environment, situation, mood, etc"? When Doom dynamically ramps the intensity of the music along with the intensity of combat, and inserts perfectly synchonized stings in time with the players actions is that not reacting to the si…
Why do you think AI music will sound like elevator music forever, when it’s already generating English text and code at such a high level? It’s quite possible that 10 years from now, Mick Gordon will sound passé when compared to the dynamic AI generated music. Maybe not, but definitely possible. There’s a lot of money to be made with better generation of music, and it’s going to be an area of exploration for sure.
Well, I would say that AIs ability to generate objectively correct text or correct code doesn't have much bearing on its ability to create worthwhile art, those are almost polar opposite goals. There is no objective metric for what constitutes good art that you can train an AI towards, the closest thing we've come up with is teaching it that the samples of art in the training set are "objectively correct" so that it will try to make something similar. Better models achive higher fidelity but are stuck forever imitating rather than exploring new or less common ideas.
Image generation is the most mature form of artistic generative AI, and the trend there has been towards introducing more human influence into the process to help guide the AI into creating something actually worthwhile. If the goal is to embed an unsupervised AI into a game engine and have it create consistently high quality and interesting music based on the current game state, with no human operator in the middle to curate and guide the process, we've got a hell of a long way to go.
Re: MusicGen: Simple and controllable music generation
#246Earlier quoted context omitted.
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…
> 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 impr…
"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.
Re: MusicGen: Simple and controllable music generation
#247Earlier quoted context omitted.
There's a real effect, just nothing even remotely close to the actual fantastical claims being made about it. It's highly doubtful there's some sort of profound way to induce arbitrary brain states through audio input alone. I remember vividly that this was very hyped in some circles around 2005 or thereabout, with wild claims that listening to some strange white noise for twenty minutes could induce full-blown psych…
I was thinking of a scenario of mapping our brain activity, like reading functions of some module, or a birthday party, or a business meeting. From then on, we put the machine to generate songs and activate roughly the same brain region of the actual life experience. We do that once, and generate 10 songs. The next time that life experience takes place, we listen five or ten minutes to the relevant songs before it ha…
Oh I see, you're essentially describing what I think of as aural contextual clues/associations. Sure, that's very real and I've experienced it first hand.
Though I'm sceptical how directed it can really be. There are some songs that have a bizarre effect on me for sure. Though most of the time it's because I had some strange experience involving the combination of said music with psychedelic drugs. And now the music can induce echoes of that experience. But it's just sort of an association that happened by accident.
I guess I could see people using this phenomenon in a more deliberate manner. And you certainly seem to be doing so. Though it could be that you're just somehow more able to than most people.
Re: MusicGen: Simple and controllable music generation
#248Style transfer is pretty cool tool as someone who likes to play around with sound design. It will be a lot of fun to drop this on parallel channels and blend it together into choruses and new instruments.
Right now the generation still sounds a lot like loop packs smashed together anyone could technically make. But it is practical for anyone who really only cares for that style of sound but do not themselves have the familiarity to do it. Now they can just say what they want and hit regenerate, skipping the latent feedback cycle of iterating with humans or sifting through song snippets.
My opinion on this style of content is that ai generation is simply accelerating us to the inevitable end of generic digital content, it isn't really changing it. It just happens to be also the optimal interface for discovering and not just generation.
Re: MusicGen: Simple and controllable music generation
#249Earlier quoted context omitted.
> 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 impr…
"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…
> 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.
Re: MusicGen: Simple and controllable music generation
#250Earlier quoted context omitted.
I was thinking of a scenario of mapping our brain activity, like reading functions of some module, or a birthday party, or a business meeting. From then on, we put the machine to generate songs and activate roughly the same brain region of the actual life experience. We do that once, and generate 10 songs. The next time that life experience takes place, we listen five or ten minutes to the relevant songs before it ha…
Oh I see, you're essentially describing what I think of as aural contextual clues/associations. Sure, that's very real and I've experienced it first hand. Though I'm sceptical how directed it can really be. There are some songs that have a bizarre effect on me for sure. Though most of the time it's because I had some strange experience involving the combination of said music with psychedelic drugs. And now the music…
That happens in general, many people associate music with relevant actions of their life. They listen to songs which are more suitable to driving a car, or lounge beats to read books.
One scenario is to record some sounds of the event once, like the laughter of a child in it's birthday, put it to songs, and listen to it before the next time it happens.
One other scenario, is to record the brainwaves of some difficult task, like programming, and by listening to songs, try to activate the same region of the brain. When there is an automatic way to create one song which activates an area but not exactly, and another song which activates one more area but not exactly, the machine will try to figure out how to combine the two songs together which will hit the spot. It is essentially a problem of combining information, which A.I. statistical engines are very good at it.