KVR Audio

jamcat · Post by **jamcat** » Tue Oct 20, 2020 11:47 pm

Listened to the 10 minute training recording of the violin. It doesn't contain any long notes. Consequently, the AI model makes all sustained notes sound like heavy handed tremolo.

audiot · Post by **audiot** » Tue Oct 20, 2020 11:59 pm

deleted

CrystalWizard · Post by **CrystalWizard** » Wed Oct 21, 2020 12:13 am

This is ML assisted resynthesis.
Tensor flow is quite fascinating, I hope that i get to see it used in some more innovative ways.
Change the thread title, there’s enough trolling on the interwebs, we don’t need it here.
Also your premise that PM is only used for imitative systhesis isn’t accurate.

Dirtgrain · Post by **Dirtgrain** » Wed Oct 21, 2020 12:46 am

I would play with it. The videos didn't have a produced sound that I thought was worthy of going in a piece (except some abstract something or other--or soak it so much reverb that you don't notice). Will it be developed more? I look forward to hearing it down the road.

perpetual3 · Post by **perpetual3** » Wed Oct 21, 2020 12:50 am

The key idea is to use simple interpretable DSP elements to create complex realistic signals by precisely controlling their many parameters. For example, a collection of linear filters and sinusoidal oscillators (DSP elements) can create the sound of a realistic violin if the frequencies and responses are tuned in just the right way. However, it is difficult to dynamically control all of these parameters by hand, which is why synthesizers with simple controls often sound unnatural and “synthetic”.

foosnark · Post by **foosnark** » Wed Oct 21, 2020 1:55 am

This isn't "resynthesis" as we are all familiar with and it has nothing to do with physical modeling. It's more like... deepfakes for music. Certainly not complete or convincing yet, but as a proof of concept, and an amusing toy, it's pretty cool.

sengoku · Post by **sengoku** » Wed Oct 21, 2020 7:57 am

but you can only turn whatever you like into a flute, violin, sax, or trumpet... what about turning a car into a bird?

E_Anderson · Post by **E_Anderson** » Wed Oct 21, 2020 8:38 am

sengoku wrote: Wed Oct 21, 2020 7:57 am but you can only turn whatever you like into a flute, violin, sax, or trumpet... what about turning a car into a bird?

you can upload your own "masks" if you want. I haven't experimented with it yet; I don't know about how deep it goes, but I am anticipating at least some really fun sound-making, for however long the novelty lasts or however deep the rabbit hole may go.

fmr · Post by **fmr** » Wed Oct 21, 2020 8:41 am

audiot wrote: Tue Oct 20, 2020 9:02 pm I could imagine that one day this technology might make other technologies obsolete, f.e. multisamples or physical moddeling, which aim on authenticity. At the moment it's hypothetic because i dont know if this technology can work in realtime.

Yes, like CDs made vinyl obsolete, digital technology made analog technology obsolete... Oh wait - they didn't

Actually, PM didn't even unveil its full potential... not even close. Even additive synthesis, which we have for like 70 years now, didn't unveil its full potential (mainly because it is difficult to control and manage). And this seems to me like some sort of additive based engine. Hardly "revolutionary".

Cinebient · Post by **Cinebient** » Wed Oct 21, 2020 12:55 pm

Google? Is that not the one still not able to offer a proper mobile OS for real-time music.
I wait for the Apple version which will come free for use in real-time inside Alchemy 3/ Logic 10.6....
or not.
Well it sounds interesting but not that great so far as it sounds in theory. Early days.

Dalle · Post by **Dalle** » Wed Oct 21, 2020 2:41 pm

audiot wrote: Tue Oct 20, 2020 9:41 pm I havent said that it's the same Physical modelling, resynthesis, scripted multisamples ... who cares. They all aim on authenticity.

Not necessarily. A lot of the fun in physical modelling is creating interesting sounds, sometimes realistic, sometimes very artificial, sometimes somewhere in between. Machine learning tends to not be that tweakable as far as I know. If you want a different result you need to train a new dataset which can be very time consuming. The point of synths for me is modifying sound in real time.

e-crooner · Post by **e-crooner** » Wed Oct 21, 2020 4:55 pm

Seems the same as with AI, tech enthusiasts seem obsessed with making humans unemployed.

audiot · Post by **audiot** » Wed Oct 21, 2020 5:36 pm

Dalle wrote: Wed Oct 21, 2020 2:41 pm
audiot wrote: Tue Oct 20, 2020 9:41 pm I havent said that it's the same Physical modelling, resynthesis, scripted multisamples ... who cares. They all aim on authenticity.
Not necessarily. A lot of the fun in physical modelling is creating interesting sounds, sometimes realistic, sometimes very artificial, sometimes somewhere in between. Machine learning tends to not be that tweakable as far as I know. If you want a different result you need to train a new dataset which can be very time consuming. The point of synths for me is modifying sound in real time.

As i said in my intro post "sorry for the sensational headline". I just wanted to generate some attention ... this is Trump style, but i think it works

Would i have clicked on a thread, that was just called "Google tone transfer"? I dont think so.

I apologize to all who love PM

My personal experiences with PM are not that convincing. Chromaphone is nice, but i prefer multisampled stuff. I can remember that i've tried to recreate a sound from Bonobo. The track is called "Cirrus" and the heart of the track is a percussive sound, that's somewhere between a bowl and a bell. From the sound demos i've listened to, i thought that Chromaphone would be perfect to get some similar results. I gave up dissapointed and used a multisampled kitchen lamp combined with a bell for my track

Reefius · Post by **Reefius** » Wed Oct 21, 2020 5:47 pm

When I see Andrew Huang I'm automatically NOT interested. That guy is annoying as f*ck and all his video's are basically ads, as long as he gets paid he will say anything you want.

highkoo · Post by **highkoo** » Wed Oct 21, 2020 5:54 pm

I gotta say, this is amazingly accessible, and sounds really really smooth.
But like everyone is noticing, this is the exact kind of example that has been used to describe next level sound synthesis for kind of a long time. "Play an apple like the color yellow!" "Make fear sound like a century!"
I have an audio sample google put out quite a while ago that must have been the early non-public experiments.

I dont think it has any technical lineage left, but this is the earliest thing I can remember that was trying to get at this kind of sound design concept;
https://www.kvraudio.com/product/plex-- ... fgang-palm

Physical modeling is dead! Google tone transfer is here ...