I watched the whole Suno Ad

There I was, watching a speculative video about what might happen to the digital music industry now that NI has been bought by inMusic.

YouTube’s algorithm says “hey, he likes making music on computers, let’s show him this ad for this thing for making music on computers!”

So the sponsor explains the new features and shows the new interface. You have to train it on your voice, so you sing a bit of a song, and because they don’t want people to use unauthorized samples i.e. someone else’s voice without permission, they make you dictate a line — just reading it in your talking voice, not singing— and they do some verification to ensure that the singing training data really was you. That seems like a reasonable thing to do, even though I imagine there are ways to get around that somehow.

So the host then proceeds to create a song that is supposed to sound like he sung it, which I thought it didn’t. I didn’t quite catch if he used AI to generate the lyrics too, and I'm not keen on trying to find the ad to confirm this, but it was a bit cheesy and could very well be AI generated.

He bops his head along satisfactorily to the soppy love song that it made for him. Despite this new version of the software — and mind you I have never used it myself — there are some odd phrasing choices and instances of falsetto which attempt to enrich the emotional content.

It’s still not very good.

I mean, of course it’s all in tune, in the way that overly autotuned singing is, with perfect pitch. He then goes on to show that you can upload more training data if you need different styles, but Suno only saves a few of these as they are used on-the-fly. You don’t keep adding new training data ad infinitum to build up a fully developed model of your singing. I guess it just uses the samples you upload in addition to all the other training data from thousands of other singers.

Next is an example of instrumental pieces, a few strings and other orchestral instruments. All sound a bit similar, kind of bland backing pieces for slideshows.

At some point was a brief glimpse of the interface that showed you have some control over the stems, like a mini-DAW. This is a good direction they they’re heading in as it means you are actually doing some of the composition. I couldn’t find a real screenshot on the main website, only this demonstrative graphic.

Come to the dark side. We have stems!

There’s a cut to another musician to tell us about something I have already forgotten, then back to the host to sum it up.

I suppose these tools could be used to let you hear what you’d sound like singing other styles that you’re unfamiliar with, like a rock song if you are a folk singer and so on. Or to make a demo that could be used to give someone else an idea of making a song with real singers and instruments. But I’ve already read comments from people resistant to that method, and they’re usually from musicians that don’t approve of the technology to begin with.

❝the songs are just too damn long and repetitive❞

It seems like that workflow isn’t happening that much, as you’re probably aware that people are just using it to make entire songs with it and no other additional overdubs, and publishing those. I have heard a small number of pieces made with software such as Suno and often there is an additional problem is that the songs are just too damn long and repetitive. The person that produced them probably used one of the older services which don’t have the DAW-like editor, so they couldn’t edit them well even if they wanted to. I’ll briefly add that I have heard one comedy song used in an Instagram video which did the job; it was amusing and felt like it was an appropriate length.

The problem then changes to: are the people using this to make songs aware that they are too long and repetitive? It’s a question of taste. If they haven’t been paying attention to what makes songs good or bad then they won’t be able to tell.

That probably is moving into music snobbery, and I don’t want to gatekeep music production. Perhaps someone who isn’t “into” music might need a song for some occasion, and they could use this to make a specific song, say for something relevant to their workplace during an otherwise silent Powerpoint slideshow about financial figures, who knows. I say this because I have seen how non-creative people have approached adding in stock photos to their slideshows for the same reason: Google image search, see the thumbnail that most closely matches their search term, then download and use it, low-res, with watermarks, no attempt to purchase the image. Visual interest added! Audience has something to look at besides the walls of text!

Yes, I know that that type of person is also not going to pay Suno or their competitors for their songmaking services, but there is a free plan offered for noncommercial releases, and there is the case of when more songmaking features move into other AI systems that are paid for by their workplace, the way image generation has sneaked in.

So if there are songs made with Suno being used in Powerpoint slideshows, I’m glad I’m not in those meetings.

🔚