from music rebalance to potential demixers, what's our future ?

[Krakatau]

Here the question his about the accuracy of some algorithms in recognizing and isolating specific sounsdsource usually :

Drums (inharmonic percussions ?)
Bass (low tessiture melodic lines ?)
vocals/choirs (human voices ?)
...and other elements in the mix

likely we can consider that these technologies are still rough but nevertheless, already effective enough to work as mastering processors

But my wonder is what can we expect as future innovations in this peculiar domain : so to say signal analyzing ,soundsource recognition and isolation ...if we can expect the arrival of a convincing demixing tool in the market in a near or less near future ?

[Krakatau]

doesn't seems wery crowded questions unfortunately but ...

...the question is how precise can be an algorithm or another in distinguish and then substract or extract a specific instrument, very likely based on his structural and timbral profile

I assume these possibilities among others :
- Guitar and plucked strings in general
- Bowed strings in general
- Electric and acoustic piano (polyphony plus sustained pedal effect)
- Wooden tuned percussion (Xylophone. marimba, chinese woodblock, etc...)
- etc...

What can we indeed, expect as better demixing algorithms coming in a near future, ...if not utopic ?

[martinjuenke]

You are here in a forum with mostly old grumpy men who love to attack each other, whine about modern developments and play Robin Hood in the audio industry. Your question is too constructive!

My opinion is:
The development of artificial intelligence is impressive but slower than their promoters tell us.
Look at Dall-E 2, the text to picture neural network. Nice little tool but far from realistic.
Look at the chat bots from Google et.al. which ended up as bad mouthed agents of conspiracy theories.
Look at the head count of self driving Teslas.

But I expect progress in all areas you mention and a lot more.
Interesting times these are...

[Krakatau]

i'm glad to hear you saying me that ...there was a time in the early moment of kvr when i could pretend to dynamise the discussions by sharing my ideas, but this seems to be past unfortunately

[kidslow]

It is an interesting question and worthy of discussion. Far more worthy than "what is your favorite X" type of threads that predominate. I wish I was more informed on the topic. To me, it does seem inevitable that the tools will improve at the same or a more rapid pace than they recently have. But when I tried demixing tools in the past I was very unimpressed. They are still in their infancy.

[midi_transmission]

You are here in a forum with mostly old grumpy men who love to attack each other, whine about modern developments and play Robin Hood in the audio industry. Your question is too constructive!

[Unaspected]

Reverb extraction - to the point where accurate IRs can be generated - would be very useful.

[midi_transmission]

For clean-up a lot of spectral processes work great. Melodyne can sound awesome when used sparse. But track separation is a very invasive process and so far all attempts sounded way too artificial to me. Humans are very sensitive to unnatural artefacts.

Will it be possible to get pristine audio quality? Only time will tell.

The question should be rather... will such technology bring real benefits? For audio repair I can see it. But I don't think that it will help to create better music.

The older I get the more I realize that we already have more tools than we need to create music. The problem isn't technology or missing possibilities.

[rod_zero]

I don't think there is enough money in the plugin industry that would allow a company to hire the developers and make the investment to develop those tools in the near feature in a way that can really displace engineers, the technology might be plausible to get as good as needed but i don't think we are close right now.

[chagzuki]

Look at Dall-E 2, the text to picture neural network. Nice little tool but far from realistic.

Eh? Dall-E and other image AIs are creating perfect blends of photoreal and stylised images in ways that human illustrators either couldn't or would take weeks.

[chagzuki]

In the case of self-driving cars the promise is probably vastly exaggerated because the systems have to be perfect, one mistake and people die. But in domains where errors don't incur such risk progress is moving at an astounding rate.

[DJ Warmonger]

Look at Dall-E 2, the text to picture neural network. Nice little tool but far from realistic.

Not everything needs to be realistic. Half of DeviantArt main page is already generated by AI:
https://www.deviantart.com/

Just so you know: synthesizing sound is way simpler problem than synthesizing the image, as it has one less dimension.

Basic commercial AI-generated music services already existed one year ago, as I checked. The only reason why these are still niche is that the market value is many times smaller than graphics, or video market.
To put it simple: The reason why AI didn't take over your music production job yet is that the market value of this job is insignificant to begin with

[Bombadil]

This de-mixer, proprietary to Peter Jackson's company, has profound implications for things from music to espionage. Anyone familiar with the original mix will immediately notice George's guitar is off to the right, rather than baked into the drums and bass track on the left.
I don't know if we'll ever see this tech commercially, but it is quite amazing.

[chagzuki]

I'm guessing that TV soundtrack production is an area where AI could make inroads very quickly.

[kidslow]

This de-mixer, proprietary to Peter Jackson's company, has profound implications for things from music to espionage.

Any good references to how that remix was done with a specific technology from Peter Jackson's company? WETA? Jackson has built something of a small tech incubator around offshoots from his filmmaking, much like George Lucas has done. I know WETA commercialized the visual tech they used to generate hordes of individually unique orcs in LoTR, and now it is used in movies ... for crowds, in stadiums and such.