Over the past ten days I've spent about as much time fiddling about with old tapes and new technology as I used to spend playing MMORPGs. Every available moment, basically.
When Beryl woke me up at a quarter to seven this morning, I was quite grumpy. I didn't go to sleep until after one in the morning last night, thanks to not being able to stop playing with my new toy.
That's the downside of having a moderately decent laptop. I haven't had enough sleep since I bought it. I was already staying up too late scouring YouTube for new music. And now this.
"This" is my current obsession: playing virtual bands. Last Wednesday, I covered the first part of the project, the one where I went through my old cassette recordings and digitized everything that wasn't too embarassing to listen to, even in private. There were a few that didn't make that cut.
After I'd transmuted the ancient tapes into MP3s, the next stage was to make them as clear and audible as possible. I was expecting that to be a problem but I was very pleasantly surprised by how well many of them had survived forty years in a cardboard box. I'd been under the impression that the magnetic tape commonly used in cheap compact cassettes in the1970s and '80s had a strong tendency to lose its integrity over time, offering up nothing much more than a few hisses and mumbles, when replayed half a century later.
It seems reports of the rate of degeneration may have been somewhat exaggerated. Most of the tapes I played sounded, there or thereabouts, as good as they ever were. Not that that would be saying much. The sound quality was always pretty poor. It still is but only one or two sounded worse than I remember.
Whatever the quality, I gave all of them at least a couple of passes through various filters and fixes in Audacity, just to be sure. At first I played around with sliders and settings trying to get the "best" sound but after a while I realized all I needed to do was amplify them and appply the default "clip fix" listed under Noise Repair. That gave me results more than clear enough for my purposes. I wasn't establishing an archive. I had other ideas.
Put simply, my plan was to hire an AI backing band to play my songs the way they'd always been meant to be played. I wasn't at all sure that was possible but it seemed like it might be. There was only one way to know for sure.
A while back, when AI was all fresh and new and exciting, in a way it really isn't any more, I spent some time playing around with a couple of the best-known AI music generators, Udio and Suno. It was an interesting experiment, as I said at the time, but it can't have been as interesting as all that because I never did much more with either of them.
I had my reasons for not following through even though, back then, I was stunned by the ability of the AIs to replicate musical genres and create whole songs from nothing more than a few prompts. True, the lyrics they came up with left an awful lot to be desired but I had plenty of lyrics of my own.
Using them, though, created as manyproblems as it fixed. It was strange, to say the least, to hear my words set to thematically similar yet significantly different melodies to those I'd always known. It felt like when you're at a funeral and you stand up to sing a familiar hymn, only to realise the organist is playing a different tune to the one you're used to. That has happened to me more than once and it's always disconcerting.
As I noted then, when I listened to AI vocalists singing my lyrics, " I can feel the new pushing out the old. I can already feel the AI singer's phrasing replacing the way I always heard it in my head." What I didn't say was that I didn't much like the sensation. That was the main reason I didn't pursue things any further.
But what if I could make the AI sing my melody instead? And adopt my phrasing? And maybe even follow my chord patterns and rhythm? What if, instead of a rough approximation, I could get a reasonable facsimile: my song but with full instrumentation and a singer who could, in fact, sing?
As I said, only couple of weeks ago I had no idea if that was possible but it seemed like it might be. If it wasn't going to cost me anything more than time to find out, why not give it a go?
Ownership, intellectual property rights and copyright, where AI is involved, is still in the digital wild west right now. Until the courts get to grips with it, which could take years, if not decades, and even then will almost certainly result in different rules for different regions and polities, it's anyone's guess who owns what. As always, whoever can afford the lawyers is going to make the rules until then.
It is fairly clear already that no-one can claim copyright of content generated wholly by AI and also that using AI in conjunction with traditional writing and recording doesn't automatically void pre-existing copyrights. Even so, who "owns" that content is far from certain. Ownership and copyright are not synonymous.
Ownership has more to do with contracts than creation. Most of the services you can use for free include clauses in the EULA making it absolutely clear they "own" what you create by using them. What ownership means in this context is that they have the right to use those songs for any purpose they like, forever. On Mars, when we get there, if they feel like it. (Pretty much the plot of Carole and Tuesday, right there.)
As far as the output of a prompt goes, I couldn't care less about any of that. The AIs can have the content they create out of thin air from my vague suggestion that they might like to have a bash at writing "a song about puppies chasing their tails in a poppy field" performed in "indiepop style, cute female vocals, toy xylophone solo."
Honestly, they're welcome to it. I just got Udio to follow that prompt and the result was abominable. Genuinely unlistenable, which is why I'm not even going to link it. I'll be comparing my recent experiences with Suno and Udio at some point in this series. Suffice it to say there's a very, very clear winner. And it's not Udio.
When it comes to actual words and music I wrote and recorded in my teens and twenties, though, I'm a lot less sanguine about handing any ownership at all over to some anonymous corporation. As in I'm not inclined to let go of any rights I might still have, regardless of whether I'll ever make use of them.
So I did a little mild research and it turns out that Suno only grabs all supposed rights for the songs you make there for free. If you give them money, that all changes:
"If you were subscribed with a Pro or Premier plan when the song was created, you are considered the owner of the song. You also retain the rights to commercial use for the song, even if you end your subscription."
As for copyright, AI affects nothing. If you owned the copyright before uploading that material, modifying it with AI leaves you with the same ownership you had at the start. In fact, you have to tick a box to say you do own all applicable rights before it will even allow you to upload anything.
I'm sure all of that can and will be challenged in the courts in due course but it's good enough for me right now. We're talking about recordings that have gathered dust under my bed since the 1990s. It's not like I was ever going to do anything with them until this opportunity arose. And it's not like I'm going to do anything meanigful with them afterwards, either, if we're going to be realistic about it.
I subscribed to Suno for a month and then immediately cancelled the subscription just so I wouldn't end up accidentally paying again. Everythng's a recurring contract, these days. It seemed exceedingly unlikely I'd need more than a month so why waste money?
Not that it's a lot. The monthly subscription is a very reasonable $10. Or £10 where I live, since apparently Sterling is at parity with the dollar now. News to me.
For that you don't just get what passes for ownership. You get 2500 credits to spend on making songs.
Suno charges 5 credits per song and always makes two songs per prompt so that comes to five hundred songs a month, which sounded like way more than I was likely to need. I mean, I only had a couple of dozen songs I was interested in doing anything with. How many tries was it going to take? Surely it wouldn't take me twenty goes per song to get something I was happy with?
Yes. Well. No. But also yes.
It didn't take me anything like that many. In fact, for the most part, I got a very satisfactory version of each song on the first or second attempt. What I hadn't bargained for was how incredibly enjoyable and addictive the whole "making different versions of my own songs" was going to be.
Like, insanely addictive. Best £10 I ever spent, quite possibly. I've used over two thousand of my credits already. Luckily, I can always buy more.
And that's about it for today. The story will continue in another post, when there may even be actual musical examples. We'll see.
Clearly this is now a series. As for the project itself, it's still ongoing. I may need a second month of Suno after all.
Notes on AI used in this post:
The big, seemingly irrelevant picture at the top is one of the images Suno automatically creates for every song. It usually pulls something out of the title or the lyrics but I have no clue why it went for hearts over a city street. I don't believe the lyrics or the title mention either.
Posts
No comments:
Post a Comment