AI Music Generator | Hacker News Reader

AI Music Generator

soundraw.io

127 pointsdeterministic37 comments

Not AI, at least not for the composition process. It algorithmically combines a library of loops using genre templates:

https://napolitano.de/2021/11/12/soundraw-ai-music-creation-...

Superb article.

To defend the website, it can still be argued to be called AI. Rule based methods are still said by some to be a section of AI.

https://en.wikipedia.org/wiki/Rule-based_system

I understand that nowadays when we think of AI we imagine deep learning and neural networks, but nowhere on the website does it claim that that is the case.

Think of ELIZA for example https://en.wikipedia.org/wiki/ELIZA

One of the earliest chatbots. Entirely rule based, but from the historical perspective undeniably classed as AI.

This is the right answer. Too many "experts" out here who cannot differentiate between AI/ML/DS but swear they know it all

It's a pretty hard problem. I'm about to run a Mechanical Turk survey comparing 10 melodies I've created together with my AI assistant to hit melodies:

https://osf.io/9nd6x

https://www.youtube.com/watch?v=51TqjKehC-g&list=PLoCzMRqh5S...

Very nice! Is the AI singing directly (producing sound waves) or does it generate MIDI that's then fed to a synth?

Something I’ve noticed when programmers try to break down melodic composition to an algorithm is that they don’t grasp how important the beat or the implied beat is to creating catchy melodic phrases. Good melodies have a question/answer or a phrase/reply relationship with this beat. It’s a conversation. All memorable music has this feature. The kick asks a question and the snare answers. The kick/snare ask a question together and the harmony answers. The harmony asks a question and the melody answers. This is what differentiates sound from music.

AI generated music has promise, but their creators have got to get this. If you’re generating a beat, harmony and melody in isolation linked only by tempo and key and then sandwich them together, you’re doing it completely wrong. They are part of one another and can’t be created in isolation or with weak interaction.

Yes. All algorithmic music I've heard (with few exceptions) feels aimless and meandering because their is no intention. It's not trying to say anything. It lacks purpose and a sense of how it fits into the greater composition. It's more like it's stringing together melodies that, while passable, don't have a feeling or target to aim for short of "stay in scale".

Melodies that I hear from programs rarely have coherency from one measure to the next, and definitely not from one section to the next. Let alone coherency with it's supporting or complimentary elements.

I've used soundraw.io and aiva.ai The first to quickly get some background music the latter to generate more original music (although harder to configure you can also edit the resulting midi directly)

It still will not replace an original composition but I can totally see a Dall-E, StableDiffusion, Midjourney moment for music on the horizon in the not too distant future

How do Soundraw know for sure the auto-generated music is original and therefore royalty-free? So not close enough to an already released, copyrighted, piece of music.

Given that not even an artist who handcrafts the whole song does know for sure I'd say they don't. They just use generic enough loops that the result is also too generic to run afoul of someone

There are 35 million released songs. Completely original music is a myth. If it doesn't use samples without a proper license, then it's legally OK.

Currently, US courts disagree.

Alot of music from previous century is just classical music playing somehow different. For example using Beethoven symphonies backwards.

The generated music didn't match my specification. Style wrong, instruments wrong.

These never have options for heavy metal :(

This should have more upvotes

Boy do I have the site for you: https://djen.co

Hm... That central element on the main page looks like it's an iframe displaying an error message.

All those AI songs sound like they are straight from sample packs previews. I am sure that in a few years we will get a real AI music generator, where you can enter a prompt like "a rock song with Freddie Mercury's vocals and lyrics about space".

Does anyone know how it works? Sounds decent, kind of generic. Maybe generated from MIDI directly?

No idea how it technically works, but from listening to a few samples it sounds like they've setup an environment with different instruments and generate notes to play those instruments.

The environment in all of the samples I've listened to so far sound exactly the same, as if they all share the same mixing configuration and mastering.

OpenAI's jukebox for example (check youtube for samples) is actually generating waveforms and sounds completely different.

The image equivalent to this is that I feel soundraw.io has a database of transparent stock photo images that it uses to compose a collage image that adheres to some strict rules depending on the genre.

Whereas OpenAI Jukebox is more like Dalle2 and Stable Diffusion.

There's a superb deep dive on the music data contained on the Soundraw site, plus intelligent surmising of its 'AI' processing, here... https://napolitano.de/2021/11/12/soundraw-ai-music-creation-...

Lol. Each Time I see such software, it's for paid. Where are open source and open data geek guys? Imagine this software or Spotify for free and open source and distributed whereas centralised.

Come on geeks

> Lol. Each Time I see such software, it's for paid

I see this a lot and this isn't open source at all.

This and others like it more closely resembles a total grift, like a selling prompts generated by stable diffusion or dalle.

> Be kind. Don't be snarky. Have curious conversation; don't cross-examine. Please don't fulminate. Please don't sneer, including at the rest of the community. Edit out swipes.

Rule #1 for comments

I didnt write this comment snarky or fulminating etc.

I would like to see an actual AI music generator with an interface based on text prompts. Soundraw seems to be algorithmic - nothing wrong with that, but they could make it clearer on their site.

By the way, this is a limitation of the paid plan:

> Sorry, you can only save 30 songs.

Edit: after trying this out for a bit, it seems like all the music is just a roughly randomized mix of the exact same loops and themes. Is anyone working on a Midjourney or Stable Diffusion for music?

Plenty of people are.

It's a much harder problem than AI art, because even tiny out-of-style mistakes in music are very jarring. We're much more forgiving of visual errors.

This reminds me of Magix Music Maker "automatic song compositor" I forgot the exact name, but It was a function inside their software that would automatically compose a song following some settings you could choose (like which genre and sample asset you could wanted). It was not AI though.

Not a music producer but I have been using this service since launch for some of my projects. Fairly user friendly I think. Although new genres have been added, older ones haven't changed much in terms of variety of melodies.

Is there anything audio generator that can do video game sound effects or foley

Sounds worse or on par with most stock/royalty free music. It's interesting that you can actually hear the absence of a soul.

I wonder if you'd have had this impression if you hadn't known that it was algorithmically created. Could be an interesting test to make, a turing test of sorts.

It’s very much like comparing AI generated human faces with faces from stock photos, you’d have to pay attention to weird artefacts and quirks. The use depends on what it needs to be used for, and for AI stuff I think it’s mostly for generic background that nobody cares too much to find those quirks.

This is anecdata, but the SoundRaw-generated audio I use on about 50% of my YouTube videos generates significantly more adverse comments than the stock audio on the remainder.

Was wondering if the generator has any tools for adding foley sounds. It'd be a nice touch, I feel

Reply on news.ycombinator.com