Best Lightweight Reverb Algorithm

[GameSmith]

I am looking for a reverb algorithm to use in an embedded app, so low cpu is required but still with the best sound possible.

I am aware of Dattoros papers on the early lexicon style revebs from 97, do they still represent the state of the art in lightweight algorithms or has something better been developed in the last 17 years?

[codec_spurt]

I am looking for a reverb algorithm to use in an embedded app, so low cpu is required but still with the best sound possible.

I am aware of Dattoros papers on the early lexicon style revebs from 97, do they still represent the state of the art in lightweight algorithms or has something better been developed in the last 17 years?

Excuse me for butting in. I've been doing a bit of research lately re: best algo vs. cpu consumption. You can find my posts easily enough by doing a search. I know nothing about DSP coding so it's just a layman's point of view.

The conclusion I came to was that 'Poor Plate' was about the best bang for your buck you could get -

This is the review by xoxos -
http://www.kvraudio.com/product/poor_pl ... nd/reviews

I have no idea what the algo is, or whether it is open source or whatever, perhaps he might know, knowing a bit more about these things.

All I know is it hardly touches the cpu (it has no gui after all which helps) and sounds very very good for a basic plate. Maybe you are looking for something more/different. But I would say, plates do the job 80 percent of the time whether it is vox/drums/gtr. A _good_ plate is a very versatile thing. In fact, I probably only do use plates 80 percent of the time.

Good luck!

[Caco]

You are welcome to give mverb a go, it's based on Dattaro's algorithm with a pre delay added in - https://github.com/martineastwood/mverb

[codec_spurt]

ss

[GameSmith]

So for low CPU it would appear to date nothing has beaten Dattaros figure 8 structure (or at least nothing published).

[MadBrain]

For low CPU, some algos have more or less the same performance as the Dattorro allpass loop:
- FDN: 8 delay matrix (with a matrix specially chosen to minimize operations)
- OpenAL algo: 4 delay matrix (chosen to minimize operations) with an allpass added to each delay (so 4 delay 4 allpass)
- Modified Dattorro: 2 delay matrix with 4 allpasses added to each delay (essentially Dattorro but with variable coupling between left and right sides)

Essentially most mixes between the Dattorro allpass loop and well designed matrix algos are probably workable.

[GameSmith]

I've tried an FDN a while back but could never get it sounding good and I could never find good information on how to choose delays lengths and feedback matrices.

I'll take a look at the OpenAL source and see what is involved.

[xoxos]

blewm uses a twice nested structure (so three allpass delays) with gain coefficients increasing "inwards" ~.25 to .75 . this feeds into the same structure routed back to itself. i thought it managed a range of settings well.

[Richard_Synapse]

The Dattoro (or rather Griesinger?) structure should be best for low CPU, since the modulation can be kept simple, and restricted to a single stereo delay. Without modulation, it's ihmo not worth bothering with low-CPU algorithms, as they just don't have enough density.

Richard

[GameSmith]

The Dattoro (or rather Griesinger?) structure should be best for low CPU, since the modulation can be kept simple, and restricted to a single stereo delay. Without modulation, it's ihmo not worth bothering with low-CPU algorithms, as they just don't have enough density.

Richard

I was wondering about modulation, that's another thing that is difficult to find information on.

Does anyone have any ideas about what works well for:

Type: triangular/sinusoidal/random?
Speed: 0.1Hz, 1Hz, 10Hz?
Depth: 1ms, 10ms, 100ms?
Interpolation: Linear/Allpass/FIR?

[Richard_Synapse]

It depends. If you modulate a single delay line only (or two for stereo), then sinusodial modulation is somewhat similar to a chorus effect, ihmo not that interesting, too predictable. If you modulate many delay lines, sinusodial modulation can work just fine, though.

Also note you can not only modulate the delay length, but also the magnitude in various ways, e.g. by modulating allpass coefficients or output taps. IIRC the very first digital reverberator, the EMT 250, used such modulation. It's still a great sounding reverb even by todays standards, just suffered from ringing, probably due to its limited amount of memory.

Richard

[MadBrain]

Type: Triangular (the "modulation rate" of sine wave is less consistent imho)
Speed: The kind of "slow chorus" speed works, so 0.1Hz up to maybe 2 or 3hz (but lower speeds work better)
Depth: I'd adjust that by ear :3 (but once again it works similar to a chorus and is kinda dependent on speed)
Interpolation: Allpass (sounds very good imho and is the way to go)

[GameSmith]

Thanks for the info, I'll give it a go and report back when I have something working.

[Ichad.c]

IIIRC the very first digital reverberator, the EMT 250, used such modulation. It's still a great sounding reverb even by todays standards, just suffered from ringing, probably due to its limited amount of memory.
Richard

That is a 'constant density' reverb, if memory serves me right, so alot of modulation would be needed to break up the repetitive pattern. Sony's reverb doesn't modulate but uses lots of delay + years of R&D, sounds way different to most verbs - very upfront.

[Richard_Synapse]

That is a 'constant density' reverb, if memory serves me right, so alot of modulation would be needed to break up the repetitive pattern.

Yep, I think it does use quite a bit of modulation.

Sony's reverb doesn't modulate but uses lots of delay + years of R&D, sounds way different to most verbs - very upfront.

Which Sony reverb?

Richard