Hey bitflipper!
First of all, thanks for your kind words! Secondly, you are writing some interesting stuff here, so I figure that I would chime in.
bitflipper
But ValhallaPlate is only a nod to EMT, not a strict emulation. If you use the Steel model, it's pretty close. But Sean took liberties with the physical properties of the imaginary plate, modifying its characteristics in unnatural ways that expand its versatility beyond what a physical plate can really do.
I feel like Cobalt is the closest to a physical plate, as it was dialed in from a specific EMT140 at Avast Recording. The rest of the plates are more "theoretical" plates.
For example, in steel the speed of sound is much faster than in air. A wavefront traverses the entire length of a plate in only ~200 microseconds. Consequently, there is no concept of "early reflections" in a plate reverb - the first reflections appear so quickly that to the ear they are instantaneous.
The weird thing about plates is that the speed of sound varies with frequency. The 200 microsecond figure you cite would be for a very high frequency. The figures that I have seen cited for audio frequencies to travel from the input transducer to the output pickups are as follows:
10 kHz: 1 msec (left pickup) 1.4 msec (right pickup)
100 Hz: 30 msec (left pickup) 44 msec (right pickup)
The sonic result of this sounds like the reverb has an instantly full echo density, but that isn't exactly true. The low frequency echoes would be kinda sparse at first, if we could perceive them this way. But we can't. A low frequency transient is really hard to hear - we tend to identify transients by high frequency energy. More importantly, by the time the first low frequency echoes come out, the high frequencies have done a few dozen laps and have turned into massive diffusion.
This speed difference between high frequencies and low frequencies is continuous, and follows a curve. I'm starting to wonder if different plates (or differently tuned plates?) follow different dispersion curves. Some of the plate impulses I heard had a really strong PEW! PEW! sound with snare drums, but the EMT140 at Avast didn't have much of that artifact. The specific dispersion curve in the modeled plates goes a long way towards emphasizing or reducing these artifacts.
I'm not quite scientifical enough to explain this concept clearly, but here goes: In a physical plate, I think that the echo density isn't really instant. However, due to the weirdness of the dispersion curve, the frequencies where we
perceive echo density get dense really quick. As long as the high and mid frequency transients are smeared out, our ear doesn't really care what is happening with the low frequencies.
Part of what differentiates ValhallaPlate's modes is their start and attack times. I'm still investigating this, but my initial observations show that the different modes implement varying delay times for initial onset of reflections, apply different attack times that determine how quickly reflections rise to full amplitude, and different release curves that affect the rate of final decay.
When developing the modes in ValhallaPlate, I wasn't thinking in terms of attack time, but rather in terms of increasing or reducing the perceived
depth from the source. This isn't a physical term, so much as a psychoacoustic one. Different plates had differing amounts of sonic depth. Some of them sounded like the source was way deep within the reverb decay, while others sounded like the source was right there in front of you, even at 100% wet.
My theory is that this perception of initial source depth is based on the phase coherence of the initial signal. If the first reflection has its phases all lined up fairly well, it will be perceived as sounding like the original signal. Scramble the phases, and things will sound like they are further away, and obscured by a lot of reflections.
This scrambling of phases will lengthen the perceived attack of the reverb. In order to diffuse a given frequency, you need to randomize its phase within +/- 180 degrees of the original input. Randomization of phases will increase the duration of an impulse by the maximum extent of the randomization. Since the period of the lowest frequencies can be around 50 msecs, this gives you an idea of how much the perceived attack will be slewed.
I think that there is also a psychoacoustic mechanism going on, where there is a time window (maybe related to masking) within which the reverb onset needs to happen, in order to sound like it is attached to the source. It is possible that this time window varies according to frequency, so that it is a shorter window for high frequency transients, and a longer window for low frequencies.
In a plate, this initial onset, and the scrambling of phases, will be different for different frequencies. So, when measuring the attack and onset time, you would want to look at this over a range of input frequencies. The easiest way to do this is to take an impulse response, and look at a spectrogram view of the output. The initial onset of the reverb will happen at slightly different times according to frequency.
There will definitely be a frequency tilt to the onset of reverb, with the high frequencies coming in before the low frequencies. This is where the "PEW!" comes from. A more randomized reverb will have a less audible "PEW!" sound, and a softer/deeper attack.
In a real plate these characteristics are based on the physics of the material and cannot be altered, so Sean imagines the characteristics of different materials, such as if the plate were made of copper instead of steel.
As you note in the followup post, the mode names are just names. All of these are based on steel plates. We tried an EMT240 in the studio, but didn't like the sound, so didn't model it. Similarly, all of these plates are "cold rolled" steel, versus the stainless steel used by the Ecoplate.
Here are some approximate measurements:
Mode Onset (ms) Attack (ms)
Chrome 20 32
Steel 12 50
Cobalt 56 60
Brass 20 23
Aluminum 20 43
Copper 14 65
Unobtainium 12 35
This is fascinating stuff. The numbers seem a little on the high side for high frequencies, but are probably accurate for some given low frequency. A given high frequency will probably have a much shorter onset, due to the dispersion curve. I never measured the attack time - I just tuned by ear, and made sure that things sounded good with various input signals and predelay settings.
A fun trick to try with the various modes, is to add predelay, and to see when you can start hearing it. For the modes you identify with shorter onsets/attacks, the predelay becomes very audible after only a few milliseconds. I think this is due to the "sharper attack" being associated with less phase randomization, so that the comb filtering of the predelay becomes audible. Meanwhile, in the Copper mode, you have to turn the predelay to a fairly high value before you can hear a difference.
Decay times vary a little with the mode, too. Steel decays 30% faster than Unobtainium even when both are set to the same Decay value. The decay curves are different as well: Cobalt drops off gradually, Chrome disappears more abruptly, and Steel has an initial bloom followed by a gradual decay.
The decay times are highly frequency dependent, and will vary according to the mode. When developing the plate models, we took measurements of a lot of plates at different decay settings, and mapped out the decay times in octave bands. From this, I calculated a table of how the decays would change in N frequency bands over time, and how the curve in between the bands would change. Next, I tweaked these a bit so they sounded better. ;)
I wouldn't be surprised if the perceived decay/frequency relationship changes with the SIZE setting. The decay/calculations are independent of the SIZE setting (i.e. the decay times should be the same, regardless of SIZE), but the sound of the decay changes a lot with SIZE. I don't know if this is due to the different modal densities, or what. At some point, theory gives way to "well, does it sound good?" The theory was more important during the initial plugin architecture, and would be revisited somewhat during the tuning process, but most of the tuning was done by ear.
Sean Costello