MTurboReverb preset making action, let's make the ultimate reverb! ;)

[Chandlerhimself]

Thank you for all the help so far. I added 2 more versions. One is percussion which I forgot to add before and the other is piano(as requested). I appreciate your honest answers, keep them coming.

https://dl.dropboxusercontent.com/u/8980900/tomhall.mp3

https://dl.dropboxusercontent.com/u/8980900/piano.mp3

[onerob]

I like reverb 1 on the toms and all the reverbs sound good on piano except 3, IMHO.

[caulixtla]

Thank you for all the help so far. I added 2 more versions. One is percussion which I forgot to add before and the other is piano(as requested). I appreciate your honest answers, keep them coming.

https://dl.dropboxusercontent.com/u/8980900/tomhall.mp3

https://dl.dropboxusercontent.com/u/8980900/piano.mp3

Toms:

1) Nice sound

2) Almost as nice, a little more flutter here — I here a distinct slapback echo

3) More metallic and slightly less clear sounding than #1

4) Significantly more muddy and somewhat more metallic sounding than #1

Piano:

1) Nice sound

2) Sounds about the same as #1 close match

3) Maybe a tad more metallic than #1

4) Less clearly defined and more metallic than #1

[Chandlerhimself]

Merry Xmas. Thank you everyone for your help. It seems people like the 1st, dislike the 2nd and feel the 3rd and 4th are ok. I brightened up the 3rd a bit and added 2 lines to the code to help reduce metal artifacts. I'm going to try to put them into an active preset soon. It should be easy and I'll let the user switch between the halls, so they can decide which one they like(hall 1 will be first though, since most here seemed to like it). Tomorrow I'll post the code for it.

@caulixtla that was an interesting guess about the topology, but I actually used a very different way. It looks like people are using a lot of different methods, so hopefully there will be a wide variety of algos.

[caulixtla]

I actually like the second one a lot, except I don’t like how it has a slapback echo to it — its attack is a little too long.

Here is my latest MTurboReverb submission, which uses ER1, LR1, LR2, and LR3, as follows:

IN>ER1>LR1>LR2>OUT

IN>LR3>OUT

ER1 is the “Chamber2” preset.

LR1, which has an input mix of 100% prev, 90% and an output mix of 100% out, 0% prev is simply three allpass filters with fairly high diffusion to smear the transients before it hits the main reverb:

a(0.7;3);a(0.7;11);a(0.7;8);

Length is 100ms because this only prepares the sound to hit the reverberant wash.

---

Next, we have the main reverb tank, LR2:

a(0.7;7);v(0.3);
fdn(8)[[fl;fh;swap];[fl;fh;swap];[fl;fh;swap];
[fl;fh;swap];[fl;fh;swap];[fl;fh;swap];[fl;fh;swap];
[fl;fh;swap]];a(0.7;9)

This is an allpass with fairly high diffusion, followed by a slightly modulated delay line, followed by the core of the reverb tank: An eight-line Feedback Delay Network (FDN) (which is modulated, see below), followed by another allpass filter with fairly high diffusion.

It took quite a bit of hits of the “seed” button before this sounded decent; the swap is there to give long reverb tails better stereo imaging.

Delay max is modulated from 99.7% to 100% using MOD1, which is a sine wave running at 0.5 hz (because the modulation range is so small, the sine wave ends up being stepped, which is a good thing here, since the stepping decreases cpu usage of the modulation; the tail still has a nice modulated sound to it, even with the stepped sine wave)

Input is 100% prev 0% in

Output is determined by a knob which I call “Attack” (the SP2016 calls this “Position”), it goes from 100% out, 100% prev (short attack using VVV parlance; “front” position using SP2016 parlance) to 100% out, 0% prev (long attack or “rear” position). In other words, a short attack means we hear the output of the input diffuser (LR1); the longer the attack, the less of the input diffuser’s output we hear, until, at 100%, we hear only the signal generated by the reverb tank (LR2).

Length is determined by a knob; it sounds good anywhere from 100ms to 60 seconds. Size is 25% by default, but can be adjusted from 5% to 60% in the easy panel. Seed, Delay Min, Delay Focus, and Delay width change depending on which one of three “room type” buttons we choose in the easy panel.

--

Finally, we use LR3 as an early reflection generator. Length is 100ms:

fl;fh;stw(32;31;30;29;28;27;26;25;24;23;
22;21;20;19;18;17;16;15;14;13;12;11;10;
9;8;7;6;5;4;3;2;1)[32a]

This is a cascade of 32 allpass filters; we make the later allpass filters quieter than earlier filters so the early reflections more seamlessly segue in to the main reverberant sound.

This gives the reverb more body, and is especially useful when used to simulate a small room or chamber, such as being used as a drum reverb.

Out is determined by the early/late knob, it can go from 100% out 50% prev (0% Early/Late mix) to 0% out 100% prev (ER3 is bypassed and not heard) when there is a 100% early/late mix — we really do not want this early reflection if making long-decay ambient soundscapes.

[MeldaProduction]

a(0.7;7);v(0.3);
fdn(8)[[fl;fh;swap];[fl;fh;swap];[fl;fh;swap];
[fl;fh;swap];[fl;fh;swap];[fl;fh;swap];[fl;fh;swap];
[fl;fh;swap]];a(0.7;9)

You a little note, this is actually the same thing:

a(0.7;7);v(0.3);fdn(8)[8[fl;fh;swap]];a(0.7;9)

And I'd improve it like this:

a(0.7;7);v(0.3);fdn[#b[fl;fh;swap]];a(0.7;9)

So the Complexity defines the size of the FDN and the dampening & swap would be enabled randomly depending on the seed, so it's sort of a polymorphic thing, which also saves the CPU. I really like the polymorphic algorithms using bypass, it lets you find things you normally wouldn't.

stw(32;31;30;29;28;27;26;25;24;23;22;21;20;19;18;17;16;15;14;13;12;11;10;9;8;7;6;5;4;3;2;1)[32a]

For this I'd recommend 2 options (to avoid the long set of numbers) :
A) Use an ER for it, it's the Diffuser mode.
B) stw(c)[32a] -> The "c" thing is new in the last secret update and it makes it use the input coefficients, so you can manipulate the weights using parameters instead of writing long sets of numbers and you can even make it follow Complexity like this:

stw(c)[#a]

[Chandlerhimself]

@caulixtla That's really interesting. I hadn't thought about using so many LR modules at the same time. I'm sure there is all sorts of crazy stuff you can come up with that way. I can't wait to hear the preset.

[caulixtla]

@caulixtla That's really interesting. I hadn't thought about using so many LR modules at the same time. I'm sure there is all sorts of crazy stuff you can come up with that way.

I got the general idea from Sean Costello (ValhallaDSP), who taught me most of what I know about reverb design: https://valhalladsp.com/2012/08/21/buil ... laubermod/

I can't wait to hear the preset.

It’s colored sounding (less so if I decrease the diffusion, but that version can sound fluttery), but comes off as sounding like an actual room.

I have come up with seven algorithms; five actually, but two of them are variants of other algorithms:

• Gravity Well and Gravity Well Sustain. This is a general purpose reverb; Gravity Well is suitable for decays up to two seconds or so long; Gravity Well Sustain changes the topology a bit, and it suitable for two to 25 seconds of reverb. Sounds a bit like Blackhole — the reverb can feel separate from the base sound — but can be used on a variety of sources.
• Golf Verb. This is the shortest topology I could find that made a useful reverb: p[#[v;9a]] Useful on most sounds; sometimes drums can flutter or sound metallic.
• Nonlin. This is my take on the “Gate reverb” effect which was very popular in the 1980s. Based on “Nonlin2” in the AMS RMX-16, and included as an algorithm in pretty much every other reverb back then. Special purpose reverb; it’s essentially pure early reflections.
• Solstice Verb. This is a fairly large set of highly diffuse allpass filters. It has a long attack, so it can’t be used for a small room, but this algorithm is good for making large ambient soundscapes using sounds with a fairly slow attack.
• Cookies N Milk. This is the general purpose reverb I posted above. Complex topology using 3 LR modules, as well as an ER module. It can be used as anything from a small very live drum room to a fictionally huge cavern with a very long sustain. Somewhat colored, so I also have a Cookies N Milk Low Diffusion variant of the algorithm with lower diffusion (allpass filter feedback): Less colored, but more fluttery sounding when used with percussive sounds.

I just spent last night debugging the active presets made for these algorithms (yes, Melda, all of them — except Nonlin which doesn’t use ERs — now allow the negative decay that was added in 10.07) and have a zip file made. I will do some more testing of them over the next couple of days before giving Melda my final submission.

Let me see if I can make a demo of all seven algorithms.

One last thing: I really appreciate all of the suggestions. I didn’t know if fdn[8[fl;fh;swap]] would work, so just made the topology longer than it needed to be to be safe. I also haven’t looked in to using the input coefficients to randomize the algorithms more. We’ll see what ideas I come up with in 2017.

[Chandlerhimself]

Here are the Algo for the clips I posted.

Hall1 -
p[#[fdn4d;b[swap;fh;fl]]];filter(12000;lowpass;0.25);fl;fh

This is a fairly basic Algo, but IMO it sounds very good. I took fdn4 and tried a few seeds until it sounded ok. The I put it in parallel and it sounded much smoother. After that I just added a low pass filter to get rid of the high end mess and other filters to let the user tune it. I used fdn4d because I think it sounds better with a denser sound. It also uses very little CPU.

Hall2 -
This isn't from MTR. It was a 3rd party reverb I used for reference. It generally sounds better, but I just chose a preset, so it doesn't sound good on some sounds.

Hall3 -
p[#fdn4];filter(12000;lowpass;0.3);b[2a;swap]

This is similar to Hall1, but there are a few differences such as the addition all passes. Since people didn't like this one quite as much I changed it to

p[#fdn4];filter(12000;lowpass;0.35);b[2a;swap;v(0.15);fa];fl;fh

The v and fa hopefully break up any resonances.

Hall4 -
r;f(12000;lowpass;0.3);b(0.5)[swap]

This is the laziest one, but it works well enough.


All of them are very CPU efficient and they work with a variety of sizes and settings. They also work without modulation, so they can be made more lush if that's what you want. I'm going to put all 3 into 1 preset, and I'm happy that they all sound fairly different.

[MeldaProduction]

Btw. instead of p[#fdn4] it's probably much better to use just "r" (or rd) in your case, as that one works in a similar fashion, but it rewires the feedback lines to avoid metallic resonances.

[caulixtla]

Here are the Algo for the clips I posted.

Hall1 -
p[#[fdn4d;b[swap;fh;fl]]];filter(12000;lowpass;0.25);fl;fh

This is a fairly basic Algo, but IMO it sounds very good. I took fdn4 and tried a few seeds until it sounded ok. The I put it in parallel and it sounded much smoother. After that I just added a low pass filter to get rid of the high end mess and other filters to let the user tune it. I used fdn4d because I think it sounds better with a denser sound. It also uses very little CPU.

Hall2 -
This isn't from MTR. It was a 3rd party reverb I used for reference. It generally sounds better, but I just chose a preset, so it doesn't sound good on some sounds.

Hall1 to my ears sounds better than Hall2. In fact, I thought it was the professionally developed preset. It has less slap back and sounds just as good while building up more quickly. I’m looking forward to having this preset available and using it with my sounds.

I didn’t like Hall3 or Hall4 as much as Hall1, but it may be good to have all three available.

A lot of reverbs don’t sound good with some sounds. The trick is to make a preset which will sound good with pretty much any sound as long as the user adjusts the controls in the easy screen appropriately.

[wesleymayhem]

I'm so psyched. When is this ultimate reverb going to be available? Thanks.

[MeldaProduction]

I'm so psyched. When is this ultimate reverb going to be available? Thanks.

Hard to say I'm afraid, my guess is second half of january.

[wesleymayhem]

Sounds great! Thanks

[Chandlerhimself]

Btw. instead of p[#fdn4] it's probably much better to use just "r" (or rd) in your case, as that one works in a similar fashion, but it rewires the feedback lines to avoid metallic resonances.

I actually did do that for the 4th algo. Using the fdn4s in parallel actually uses about half the amount of CPU and normally I'd just switch, but it seems people here(and myself) liked the sound of the fdn4 as it is, so I decided to stick with it. I figure if I change it I might screw it up or start tweaking in circles. If it isn't broke, don't fix it.