Reaktor - schroeder allpas

gentleclockdivider

Thanks guys , much appreciated

gentleclockdivider

I've got another question
I am comparing the pure data allpass~( cyclone library ) to reaktor's core allpas .
Reaktor core allpas , 2 delays in series

IN total , I ve go 6 allpasses in series (two in each core cell) , all with the same gain coefficient of 0.5 .

The allpas frequency is samps to s ( to ms for guireadout )

BOth are identical when it comes to frequency and gain coeff .

When feeding an impulse in both , there are some small differences
Top is pure data , bottom is reaktor

When zoomed in , in pure data the ouptut are only impulses of 1 sample long , in reaktor there is an obvious variation

colB

https://community.native-instruments.com/discussion/comment/144166#Comment_144166

What does the impulse look like, is it the same for both, and how are you feeding it in?

Can't help you with the pure data stuff, but you will need to ensure that the pure data code means exactly the same thing as the Reaktor code.

And sanity check the delay times after everything has ben converted to number of samples inside the process. Also double check they are running at the same sample rate and stuff like that.

try sending the impulse through both without processing to see what that looks like…

gentleclockdivider

You know what the problem was ?
IN reaktor I played the impulse file (1 unit sample spike ) one octave lower , ad thus renoise interpolated the impulse impulse before going into the allpasses , which had a totaly different outcome (obviously) .

gentleclockdivider

What an absolute rabbit hole reverb designing is .

Started of with some combs and allpasse in series , basic schroeder stuff untill I've read about nested allpasses

.A series of allpasses inside a delay that is also allpased , I wonder where exactly to put it .

Here is a core macro of 3 allpasses in series , don't know where exactly to put it inside the master allpass delay

I think the green lines are the best option

But the diffusion only starts after a first loop , making the first loop dry

Which I think is correct , when taking this structure as an example ( from vallhalla website )

colB

https://community.native-instruments.com/discussion/comment/144683#Comment_144683

The paper you need to look at is:

"The Virtual Acoustic Room" by William Grant Gardner

https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=7c3e3f3947d49aa8aaf561ca1d138aee9e93f1f3

This has diagrams of nested all-pass structures for small, medium and large rooms. They are pretty decent sounding too.

The other is

"Effect Design: part 1, reverberators and other filters" by Jon Dattorro

https://ccrma.stanford.edu/~dattorro/EffectDesignPart1.pdf

This has a diagram for the classic Dattorro plate near the start. It's a nice sounding reverb for low cpu, but maybe not as closely aligned with what you are working through...

Both papers also have lots of juicy analysis, worth wading through, even if some of the math is just too dense to follow. The Gardner paper is much less mathy, and so easier to follow, it's really a great resource.

(I'm not sure why the forum software turns these links into crazy gibberish, but... I think they will work. Otherwise, it's easy enough to find these papers in various places)

Main thing is to try tweaking these models, not just creating a direct copy and leaving it at that....

FWIW, my favourite reverb approach so far has been FDN, you can get really smooth unrealistically dense verbs with this. Great for synths.

gentleclockdivider

Reading a paper by griesinger , and there the schroeder all pass negative coefficient is taken before the sum with obviously a different result .

gentleclockdivider

I don't just copy paste stuff but some beginner guidelines are verry welcome .

Ofcourse I just started with comb into allpass etc..but there is nothing worse than thinking you know stuff while in fact you don't and that's where the confusion begins .

Diagrams of allpasses ,( see previous post) where the G.C. is taken before the sum etc...

JUst doing it for a few days now , and the best result I got is 5-6 allpass in series inside an allpass delay .

Some allpasses to smear the transient too .

Either way , it's not my goal to create a wonderfull reverb but the subject is verry interesting .

I' m afraid the math involved for FDN is way above my head

colB

https://community.native-instruments.com/discussion/comment/144834#Comment_144834

I' m afraid the math involved for FDN is way above my head

me too, but I still got some good results, and not sounding the same as everyone else's

I find the shroeder allpass thing to be very difficult to tune. The Gardner examples are great, and they show just how good it can be, but how he managed to achieve that is something else. It's a real art to create a system like that and tune it so that it doesn't just sound like a metallic phasing mess.

Maybe I could get a better result now from the things I've learned through messing about with FDN structures.

The Dattorro one sounds great, but it is more homogenous, so there is maybe less opportunity for coming up with something noticeably original from it?

gentleclockdivider

Yep , the dattorro digital plate is pretty easy , at least the strucuture shown here ( lot's of missing stufff )

It's basically the same as the alesis midiverb algo Pogram 22 .

I made some ready to go macro's to quicky replicate the structure , now replacing the 4 allpasses on top with a 6 nested alpasses in an allpassdelay will give you some huge sounds .

Alough I do think that mst of the structures found on the web are bassically nested allpasses without mentioning it .

The decay part in the tank section is confusing , it's obviously a gain coefficient to control the overall feedback but it doesn't really make sense to tap the output from there to dac , instead I tapped it from the delay in front of it

Edit..just as I thought the dattorro taps from multiple points in the tank to output ,

colB

Yes, the crucial thing about Dattorro is that figure 8 feedback. The Alesis diagram doesn't do that, but the last one in your post looks closer. Here's the diagram from the paper:

There are more details in the paper about parameters and output taps.

gentleclockdivider

What's interesting about the midiverb is that taps are taken pre and post each delay line , these are then summed for l/r

According to the guy in the video , this strucuture was used for most of the midiverb algoritms

Also all diffuser gain coefficients are 0.5

https://www.youtube.com/watch?v=5DYbirWuBaU

Here is an attemp ( it does sound kind of ****** :) ,

I have the uberverb vst which is an algoritmic midiverb emulation ( most of the diagrams are to be found on gearslutz written down by an allround genius guy called acreil ) then another guy made a vst out of it :)

midiverb.ens

gentleclockdivider

This video is gold , lecture by the great tom Erbe ( soundhack etc...)

https://www.youtube.com/watch?v=Il_qdtQKnqk

gentleclockdivider

Here's the ensemble with the midiverb , made in core .

I just have a hard believing that these limited set of parameters (allpas coefficients and delay values ) are capable of the midiverb sounds , at least that's NOT what is happening in the ensemble .

IIRC midiverb had a maximum delay length of 3 sec ,any help welcome

It sound mediocre to say the least

WHat's also weird , while modelling the tank which needs feedback , when set to solid the orange feedback feedback lines dissapeared , but now they are always orange regarless solid setting or not

MVcore.ens

colB

https://community.native-instruments.com/discussion/comment/145223#Comment_145223

I just have a hard believing that these limited set of parameters (allpas coefficients and delay values ) are capable of the midiverb sounds , at least that's NOT what is happening in the ensemble .

Do you have the coefficients and delay times they used? thats where the goodness is for sure. And that's the difficult thing to get right, making reverbs is an art, and most of the art is in choosing these values.

EDIT: I also notice that the diagram looks to me like the outputs are from multiple taps in each of the delay lines... and that there is no specification for those all-pass blocks in either the diffusion section or in the feedback loop. There are so many different things that are technically an all-pass. A delay line is an all-pass. A 20 pole multi-layered nested all-pass is also an all-pass.