Reverb

[camsr]

A dynamic convolution network for reverb would be quite interesting.

[karrikuh]

Dense pseudo-random reverb tail is easy to get: use many parallel Hadamard matrix FDNs at equal output gain. Each FDN is initalized with a random delayline length. Produces perfect white noise tail. Nothing to look for beyond this, can't get any better reverb tail.

Can you give ballpark numbers which size those FDNs should be at the least and how many of them you would need in parallel to get an uncolored tail?

[Aleksey Vaneev]

Dense pseudo-random reverb tail is easy to get: use many parallel Hadamard matrix FDNs at equal output gain. Each FDN is initalized with a random delayline length. Produces perfect white noise tail. Nothing to look for beyond this, can't get any better reverb tail.

Can you give ballpark numbers which size those FDNs should be at the least and how many of them you would need in parallel to get an uncolored tail?

I'm using 4x4 Hadamard matrix FDNs, with absolutely all delay line lengths at random in the range 5 to 100ms. I run 16 of them in parallel with the same input signal. So there is a total of 16*4 delay lines of random lengths. The number of FDNs can be reduced I think, but even at that number the CPU load is quite low. Used random delay length range can be reduced e.g. to 30ms, but this will produce a "plate reverb" character.

[nonnaci]

The room reflection buildup can be exactly computed in special cases (integer sized shoebox type rooms) without raytracing methods to create an FIR as it can be expressed as a dynamic programming problem. I basically did this in my reverb and generalized the math to all higher-dimensions shoe-box type rooms. See my blog/tutorials in sig. for derivation.

[Richard_Synapse]

The room reflection buildup can be exactly computed in special cases (integer sized shoebox type rooms) without raytracing methods to create an FIR as it can be expressed as a dynamic programming problem. I basically did this in my reverb and generalized the math to all higher-dimensions shoe-box type rooms. See my blog/tutorials in sig. for derivation.

Does the reflection buildup actually exist? I'm just wondering because I've never seen any impulse response of a real room or hall that consisted of more than just some form of noise, plus sometimes a few distinct early echoes buried in that noise.

Most algorithmic reverbs build up over time because that's what naturally happens when concatenating delays (and it sounds great that way, so this is not a question about sound quality), I just doubt that this is what happens in any real space, because the buildup seems way too fast - unless it is the measurement methods when recording impulse responses which produce too much noise or something.

Richard

[nonnaci]

Does the reflection buildup actually exist? I'm just wondering because I've never seen any impulse response of a real room or hall that consisted of more than just some form of noise, plus sometimes a few distinct early echoes buried in that noise.

Most algorithmic reverbs build up over time because that's what naturally happens when concatenating delays (and it sounds great that way, so this is not a question about sound quality), I just doubt that this is what happens in any real space, because the buildup seems way too fast - unless it is the measurement methods when recording impulse responses which produce too much noise or something.

Richard

The buildup depends on size of room and the amount of material absorption coefficients of the walls. Since most materials aren't plexi glass or something with super-low absorption and rooms aren't large enough to space out higher-order reflections, the resulting RT60 is short. In simulation tho, these conditions can be easily met and what you get are these long-tail reverbs.

EDIT: Also, any objects in a real room tend to act as diffuses which further lower RT60. In an idealized simulated room where certain physical properties can be relaxed (no diffusers, ignore non-reflection of wavelengths greater than size of wall), different effects are possible. This isn't a failure of modeling insomuch that the laws abstracted from the real world hint at new possibilities or "interesting spaces" only achievable in simulation; e.g. no tesseract exists even approximately in real-world but from the few laws hinted at in 3D world, we can extrapolate what an experience would be like.

[camsr]

Most algorithmic reverbs build up over time because that's what naturally happens when concatenating delays (and it sounds great that way, so this is not a question about sound quality), I just doubt that this is what happens in any real space, because the buildup seems way too fast

An orthogonal 6-sided rectangular room fits your description. But what can de-emphasize the buildup of the diffuse soundfield is actually not related to the room at all; the directivity of the sound source has a major influence on the resulting room echo. A perfect, real life example is the tweeters on many speaker cabinets, these are highly directional towards ultrasonic frequencies, and as such do not have the same radiation pattern as the same cabinet's woofer. This also means that even if a room is strongly reflective, these soundwaves are initiating a completely different trajectory back to the listener.
IR recording can capture this 3d effect of a source sound in a room, the limitation being only one position is isolated. There's no information about the rest of the room or any possibility of what occurs inside other spaces in the same room.

[nonnaci]

An orthogonal 6-sided rectangular room fits your description. But what can de-emphasize the buildup of the diffuse soundfield is actually not related to the room at all; the directivity of the sound source has a major influence on the resulting room echo. A perfect, real life example is the tweeters on many speaker cabinets, these are highly directional towards ultrasonic frequencies, and as such do not have the same radiation pattern as the same cabinet's woofer. This also means that even if a room is strongly reflective, these soundwaves are initiating a completely different trajectory back to the listener.
IR recording can capture this 3d effect of a source sound in a room, the limitation being only one position is isolated. There's no information about the rest of the room or any possibility of what occurs inside other spaces in the same room.

Directivity and relative positioning of emitter/receiver pairs in the space are indeed important but for simplicity in modeling, most sound-sources are treated as point-sources with equal pressure in all directions. The effects of positioning has more an effect on the spacing between the direct and early reflections before any buildup (if any) kicks in + frequency dependent room modes (the latter of which may be undersirable). Maybe look at some IR datasets that have multiple measurements taken within the same space.

[camsr]

Directivity and relative positioning of emitter/receiver pairs in the space are indeed important but for simplicity in modeling, most sound-sources are treated as point-sources with equal pressure in all directions. The effects of positioning has more an effect on the spacing between the direct and early reflections before any buildup (if any) kicks in + frequency dependent room modes (the latter of which may be undersirable). Maybe look at some IR datasets that have multiple measurements taken within the same space.

Omnidirectional sound radiation is a special case, there's not many things that naturally do this.

[nonnaci]

Omnidirectional sound radiation is a special case, there's not many things that naturally do this.

That's true despite how IRs used to be measured with popping balloons. Speaking of which, has anyone thought of making a speaker from attaching a driver to an air-compressor and a balloon?

[JCJR]

Speaking of which, has anyone thought of making a speaker from attaching a driver to an air-compressor and a balloon?

I can't recall names or links, but there have been very-low-frequency subwoofers intended to be built-in to rooms, which use electromechanical rotary drivers which more-resemble servo-controlled air-compressors than conventional speaker drivers. The installs I saw online, as best I recall, looked like using an empty room adjacent to the listening room as a "big woofer cabinet", with the rotary device pumping air between the two rooms. Kinda like industrial-strength wall-mounted woofers.

In general most all of the bass in a room can be considered omni source. OTOH dunno how many people DESIRE much reverb on bass. I considered reverb on bass useless, merely adding mud to the mix. I always tried to keep the low end quite dry. But some music seems to bury kick drums in reverb, so maybe some folks also like to bury bass guitar or synth bass in reverb. Not to my taste anyway.

In other words, maybe I'd sometimes want to reverb a synth bass, but the reverb would be hipassed so that only the higher harmonics of the bass get reverb.

[aciddose]

Does the reflection buildup actually exist? I'm just wondering because I've never seen any impulse response of a real room or hall that consisted of more than just some form of noise, plus sometimes a few distinct early echoes buried in that noise.

This is the issue with mirror-like reflections vs. reality where diffusion is far more prevalent. Rather than a single echo you get more of a splatter.

Yes though, mirror-like reflections do exist off very solid materials like stone cliffs and lake surfaces. Have you ever heard some loons in a mountain lake?

(Bad example, I couldn't find anything even close to how awesome it sounds of 50 ft cliff faces in mountainous BC terrain)
https://www.youtube.com/watch?v=4ENNzjy8QjU
You can clearly hear the distinct echos although the delay time is significant due to the distances somewhere around 300 - 500 meters. ("Long mournful wail" which means "RU DTF?")

https://www.youtube.com/watch?v=t2iBV1oM65s
edit: Better one.

Unfortunately approximating diffuse reflection is far more intensive and can't easily be accomplished with discrete delay lines.

A complex all-pass matrix works but we're talking about 1000s of filters to approximate each wall reasonably well.

[camsr]

In general most all of the bass in a room can be considered omni source. OTOH dunno how many people DESIRE much reverb on bass. I considered reverb on bass useless, merely adding mud to the mix. I always tried to keep the low end quite dry. But some music seems to bury kick drums in reverb, so maybe some folks also like to bury bass guitar or synth bass in reverb. Not to my taste anyway.

In headphones, LF reverb can be acceptable and if leveled in properly, not muddy. In most small sized rooms, we never hear VLF reverberation, because the wavelengths are too large. But we do hear the effect of contained LF in these spaces as either boosts or peaks in the FR, and possibly resonance.

[xoxos]

(Bad example, I couldn't find anything even close to how awesome it sounds of 50 ft cliff faces in mountainous BC terrain)

north of marana, arizona, crystal visions road area. think of a pinball table twenty miles across lined with cliffs or steep ranges. rancher fired a rifle on the other side of an isolated ridge like a bumper in the middle of the table.

i've tried to describe that here before.

[harryupbabble]

Hay, doesn't Stevie Nicks live in Arizona? I bet that Crystal Visions Road is named after a lyrics line in her Dreams song.