KVR Audio

Winstontaneous · Post by **Winstontaneous** » Tue Oct 20, 2020 8:53 pm

Act fast - before:
- Google is split up and they can't funnel ad/search revenue into R&D/side projects: https://apnews.com/article/google-justi ... 5d4db06044
- Google discontinues cool technologies they've made that people become accustomed to, just 'cuz: https://killedbygoogle.com/

Scotty · Post by **Scotty** » Tue Oct 20, 2020 8:58 pm

This is cool ... it isn't new... We've had resynthesis for a long time. It isn't a replacement for physically modeled synths which seek to replicate the mechanics of wave motion, membranes, strings and tubes. It will be useful regardless.

audiot · Post by **audiot** » Tue Oct 20, 2020 9:02 pm

nathanj wrote: Tue Oct 20, 2020 8:45 pm Maybe just change the name by removing physical modelling. It’s a bit misleading.

And you get people like me who love physical modelling coming in chests puffed ready for a brawl

Thanks for your suggestion, but i'm currently in the mood to piss off some physical modelling fanboys

I've been way too nice lately ... yin and yang, you know

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.

onerob · Post by **onerob** » Tue Oct 20, 2020 9:05 pm

https://sites.research.google/tonetransfer

nathanj · Post by **nathanj** » Tue Oct 20, 2020 9:11 pm

But it is nothing like physical modelling. This is Resynthesis.

It's like saying FM is Dead. Granular synthesis is here!!!

Unaspected · Post by **Unaspected** » Tue Oct 20, 2020 9:38 pm

audiot wrote: Tue Oct 20, 2020 8:15 pm Maybe someone who knows a bit about DSP stuff, could give an opinion on wether this will be available in realtime soon? I'm really curious ...

When NPUs are commonplace we might shave a few hours off.

The audio engine is comprised of three main modules: Partials; Noise; Reverb/Filter. The machine learns how best to tune these modules over the course of several hours until it arrives at the best match to the intended sound. It is educated trial and error.

This video gives a good visual representation of what is happening without going deep into how neural networks operate: But the basic concept is more simple than the name might suggest. I've yet to jump in but if you understand DSP it certainly opens the door.

audiot · Post by **audiot** » Tue Oct 20, 2020 9:41 pm

nathanj wrote: Tue Oct 20, 2020 9:11 pm But it is nothing like physical modelling. This is Resynthesis.

It's like saying FM is Dead. Granular synthesis is here!!!

I havent said that it's the same

Physical modelling, resynthesis, scripted multisamples ... who cares. They all aim on authenticity. Will battery cars or hydrogen engines replace the combustion engine? Maybe! It's just a sensational thread title ... it's click bait

I dont want to discuss technical details. What i've heard in the clips, was a new level of detail in the articulations and i thought it would be nice to share my enthusiasm

And sorry for the car allegory

SparkySpark · Post by **SparkySpark** » Tue Oct 20, 2020 9:54 pm

audiot wrote: Tue Oct 20, 2020 8:43 pm Oh ... 2 downvotes already. Sorry for bringing this thread up. I promise that my next topic will be something like "Synth X vs Synth Y" ... or something like that

Thanks for bringing this up!

(Besides, I predicted PM would have made a mark several years ago, but it never did, so there's that. I had high hopes for my virtual twenty metre carbon flute, but oh well...)

V0RT3X · Post by **V0RT3X** » Tue Oct 20, 2020 9:57 pm

Interesting.

Looks like a good way to profile tones, so maybe they could use this along with PM to make it more convincing.

audiot · Post by **audiot** » Tue Oct 20, 2020 10:00 pm

Unaspected wrote: Tue Oct 20, 2020 9:38 pm
audiot wrote: Tue Oct 20, 2020 8:15 pm Maybe someone who knows a bit about DSP stuff, could give an opinion on wether this will be available in realtime soon? I'm really curious ...
When NPUs are commonplace we might shave a few hours off.

The audio engine is comprised of three main modules: Partial; Noise; Reverb/Filter. The machine learns how best to tune these modules over the course of several hours until it arrives at the best match to the intended sound. It is educated trial and error.

This video gives a good visual representation of what is happening without going deep into how neural networks operate: But the basic concept is more simple than the name might suggest. I've yet to jump in but if you understand DSP it certainly opens the door.

Interesting

I've also watched a few videos about tensorflow.js a few months ago, but i didnt had the time to go deeper.

A few music companies like Sonible are already experimenting with AI ... although i sometimes have the feeling, that AI is more of a marketing term to some companies. Let's see!

hotmitts · Post by **hotmitts** » Tue Oct 20, 2020 10:01 pm

Nietzsche is also dead, which incidentally is a bugger to spell

SparkySpark · Post by **SparkySpark** » Tue Oct 20, 2020 10:39 pm

hotmitts wrote: Tue Oct 20, 2020 10:01 pm Nietzsche is also dead, which incidentally is a bugger to spell

Well, maybe his first name was PM.

SeeingInMidi · Post by **SeeingInMidi** » Tue Oct 20, 2020 11:27 pm

I remember reading about this project a few years ago when it was named something else. Nice to see that they're still working on it.

Machine learning is the future. Can't wait to see what the future holds for AI oriented DSP. I've been saying this for a while ever since Zynaptiq and iZotope have hopped onto machine learning.

felis · Post by **felis** » Tue Oct 20, 2020 11:36 pm

vurt wrote: Tue Oct 20, 2020 7:54 pm tell me in words

cloud crap

audiot · Post by **audiot** » Tue Oct 20, 2020 11:46 pm

hotmitts wrote: Tue Oct 20, 2020 10:01 pm Nietzsche is also dead, which incidentally is a bugger to spell

Friedrich Nietzsche is overrated! Oprah > Nietzsche

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