AI start-up Suno AI from Cambridge generates lifelike radio-quality music

Here’s an AI-generated music of a sad girl singing the MIT License:

Audio Thumbnail

The road not taken:

Audio Thumbnail

Another one:

13 Likes

This is INSANE.

Just did 2 lines. The app is a bit understandably buggy but … wow…

3 Likes

After listening to several dozen of the songs it became obvious there are algorithmic templates under the hood. It’s still impressive, but far from covering a wide range of musical expressions and styles. Pop music is simple by definition.

Music is by many orders of magintude more complex than pictures and language. The dimensionality is brutal.

1 Like

Somewhat agreed. My girlfriend and I played around with these for a while and a lot of the results sounded like cheap, commercial fiverr-like songs. Maybe a consequence of using royalty-free music

There are some hidden gems though. I’m currently paying $10/month for 500 songs ($0.02 / ~1 min song), which is incredible.

The amazing part of this service is the ability to use lyrics. It supports outpainting, and soon v4 will support inpainting.

Of course, artists will always outshine. There are some songs I have tried to replicate and the engine just simply cannot grasp it

1 Like

The platform will have a hard time making profit. Consumers don’t have much use for such a tool. There is music on demand in abundance for every conceivable taste and purpose, e.g. on Spotify. Why bother making your own?

Musicians may feel compelled to use it for inspiration, but only for 1-2 months in a row. The niche market of music production and bands is incredibly small.

Great share!

I thought that is the current secret recipe for AI tech in general? Find something that works well and then scale it. See what else emerges. But I agree that using these tools does have a district impact on style.

Definitely! And it’s super user friendly. I love this part the most.

But businesses do!
Finally I can whip up the ‘super tight’ rap for my new series of metal sealings. Of course the professionals who today spend parts of their days finding the right stock material will now move on to create a regularly matching background sound but I guess the musicians creating the original sounds will feel the pinch.

3 Likes

Copyright.

If I want some music for my YouTube video or my advert … who can complain about royalty free?

5 Likes

I made a song for my son about him and his best friend (his sloth stuffy) going on an adventure in a rocketship. So, there’s that.

Then you’ve got the market for license-free music, people who want to put background music in their projects (films, commercials, etc).

You’re also underestimating the market of people who want to create music but cannot play an instrument or sing.

So, I think these companies will do just fine.

4 Likes

This is really good. It blew my expectations. It is one of my dream to make AI sing the texts you provide them. I just made one using the Gettysburg Address into 80s pop rock song (imagine Scorpion singing Wind of Change) and I can feel the hair of my skin standing.

4 Likes

Sure this is fun and everyone wants to play with it for a while, do a song for friends and family, but then it gets old and repetitive. All the same pop clichés over and over. I don’t see why anyone would subscribe to the service for years on end. This is not Netflix.

If you need royalty free music for videos, blogs, games, ads, business, etc, you want something more specific with better (instrumental) quality. Note some 1970’s transistor radio sound. With a stock music library you at least get a finished product. With suno you need to reproduce the song in a (home) studio, which is a lot of work with uncertain outcome.

A “market of people who want to create music but cannot play an instrument or sing” doesn’t exist. There’s one market of people that have fun creating music themselves (whether they actually can or not) and another market of people that want to feel like they have created music.

The first market is small but sustainable (thousands $$$ are spent on equipment, courses, instruments, software). The second market is big, but with a huge churn (they might try it once or a couple times, then go on to do other things).

It seems like someone hasn’t been paying much attention lately.

This is the worst it will ever be. If you think it won’t be an order of magnitude better in a year, you should look around a bit.

4 Likes

It is certainly buggy but it is also a great proof of concept that AI can generate good sounding piece of music, with great sounding vocals.

As @anon22939549 points out that things only get better from here, there’s certainly a huge market for them, which is only going to grow as time progresses.

1 Like

In response to this part of your post I would point you toward the “no-code developer” space.

There’s a lot of money to be made serving those whose vision exceeds their reach. Anything which can credibly promise to bridge that gap will be immensely valuable—no matter how you, I, or anyone else feels about it.

raises hand

Who doesn’t want to express themselves creatively? There is a Discord channel with loads of people pumping out and sharing music that they “created”, and it sounds awesome.

Last night I prepared a italian dinner with my girlfriend and we generated Italian music to accompany us. It was GREAT. I would definitely pay… ~$0.02/minute or ~$1.2/hour to generate live, tailored music.

Eventually I bet I could create a feedback loop where the generated music runs through another AI to grade it, then once it’s good it can become a part of a playlist for my house/restaurant/whatever

This is the same song and dance with image artists, programmers, vocalists, writers, etc. As a programmer it’s a scary time but it’s also so damn interesting and exciting

4 Likes

Ok. What you are talking about is not “creating” music. You don’t create anything. You just ask an AI to come up with something. It’s random radio.

I’ve been in music production for decades. It takes years of training and painful setbacks and failures before one is able to produce something that passes as good enough to work for a mainstream audience. Just the engineering (songwriting not even included). Listening takes years of training too. Most listeners can’t even tell the instruments apart.

The suno songs are music in the same way that McDonalds is a restaurant.

1 Like

It’s all relative.

I could take Taylor Swift and raise you Angie Stone, or Liane Carroll.

It’s not like modern music production doesn’t use lots of tricks and shortcuts already.

Autotune, quantisation, cut & paste, fake drum kits, samples, physical modelling even!

This is just another production process … but now potentially fully automated.

And I’m not sure I expect it to be very good, not for a long while …

… and nothing is going to replace the thrill and emotion of the live performance and witnessing raw human talent first hand.

Orders of magnitude. That’s exactly the hurdle that can’t be overcome. There is no labeling that people could agree upon and no embedding that could cover the high dimensionality of music. The depth is insane.

That’s why they resort to templates and formulas. They will make new algorithms/formulas for more styles. But essentially algorithm is style. So every algorithm can only produce a single style.

Agreed. Unfortunately most people like McDonald’s. Hits the spot at 3am.

There are still really delicious, outperforming, timeless restaurants though.

Holy! what the? I would actually buy that track… dayum

2 Likes

This is actually amazing, but this topic is starting to sound like;

Let’s not have that happen, we are a developer forum after all, so here’s their GitHub:

Have fun :wink:

5 Likes