IRs made with tube amps (ML Sound Lab claims)

[Orvillain]

Actually, it doesn't, and it can't.

Do you mean purely in the context of IR loaders?

 

[jay mitchell]

what about condenser mics that have a figure of 8 pattern?

There's no such thing. Ribbon mics are figure of eight, and they must be dynamic. There's no practical way to make a condenser ribbon mic.

 

[jay mitchell]

And it’s all the more reason for mics to retain their original phase response.

You also need to be aware that, in a travelling acoustic wave, particle velocity, and acoustic pressure are in phase with each other. Ergo, pressure- and velocity-sensing mics are not intrinsically 90 degrees out of phase. The aiming of a directional mic affects the phase of the signal it picks up - for example, the rear lobe of a dipole (figure of eight ribbon) is 180 degrees out of phase with the forward lobe.

 

[jay mitchell]

Do you mean purely in the context of IR loaders?

No. It's physically impossible.

 

[MirrorProfiles]

There's no such thing. Ribbon mics are figure of eight, and they must be dynamic. There's no practical way to make a condenser ribbon mic.

Maybe there’s some confusion here. You absolutely can set a C414 or U87 or U67 etc to figure of 8. Those are absolutely not ribbon mics, obviously.

I got the pressure/pressure gradient thing from this video, which someone corrected (with a similar point to yours) in the comments.

 

[Orvillain]

No. It's physically impossible.

I don't understand. You seemed to be saying interpolation doesn't and can't work well accuracy-wise, and I was looking for whether you mean specifically in the domain of blending IR's together, or as a mathematical operation in general. Because my understanding is that more or less without interpolation, digital audio simply doesn't work.

To my knowledge, in the realm of IR blending, you've got time domain interpolation and you've got spectral domain interpolation. Time domain is cheap, but doesn't sound very good. Spectral domain is expensive and complicated to make work in real-time, but can sound decent depending on how the system is tuned.

I built a prototype that allows me to switch between the following methods:
Time domain trilinear blending (multiplying the weights)
Spectral minimum phase blend - blending magnitudes, discarding the original phase information, reconstruct phase via cepstrum
Phase unweapped spectral blend - blends both magnitude and phase data; phase unwrapped before interpolation.
Optimal transport blend - treat magnitude spectra as probability distributions; compute a Wassersteiner barycentre, and then rebuild a minimum-phase IR from that

The basic tests I've done with them have definitely resulted in different tonalities; and it left me wondering whether that is why I prefer some of these IR blenders over others.

It was quite hard to choose a preference though, because I'm working with a less than ideal data set currently. I borrowed a Dyna-Mount from @MirrorProfiles (thanks bud!) in order to take a finer grain set of impulses, but I just haven't had chance yet!

 

[jay mitchell]

I don't understand. You seemed to be saying interpolation doesn't and can't work well accuracy-wise,

I'm saying that interpolation between two IRs taken at different angles won't yield the same result as an IR that is captured at the desired intermediate angle. Apps that allow you to alter virtual mic placement rely on a collection of IRs and interpolate between two or more to provide an IR for an intermediate position at which an actual IR was not taken. This is physically impossible. The interpolated IR may be pleasing - a qualifier I'm forced to always add - but it will not be the same as an actual IR that was captured at the intermediate position. This is a key issue for sound system modeling/design software, and it is widely ignored by people who should know better.

and I was looking for whether you mean specifically in the domain of blending IR's together, or as a mathematical operation in general.

Interpolation (and mathematical extension beyond collected data) is a useful technique, but it necessarily relies on assumptions about the physical system in the model. When those assumptions are not met - as in the case of IR interpolation - the result will be incorrect.

The basic tests I've done with them have definitely resulted in different tonalities; and it left me wondering whether that is why I prefer some of these IR blenders over others.

Mixing IRs is not at issue here. It's predicting what an IR you did not acquire would be using data from neighboring positions. That is physically impossible in a general sense. If it occasionally appears to work, it's random chance, not good practice.

The underlying principles involve projection of a four-dimensional (x, y, z, and t) space into a two-dimensional (v [signal magnitude] and t) one. They apply to analog as well as digital electronics. The four-dimensional space contains information that can't be definitively projected into the two-dimensional one.

 

[jay mitchell]

Maybe there’s some confusion here. You absolutely can set a C414 or U87 or U67 etc to figure of 8.

That was my oversight. Mics that use more than one capsule can indeed have a dipole pattern without being ribbon types.

 

[FuzzyAce]

It doesn't really matter what you use (solid-state, tube) as long as you capture the actual voltage at the speaker terminals and deconvolve the measured response with the measured voltage.

 

[riffy]

OK, I’ll bite. You guys tell me if I’m completely goofy or not, and if I am, it won’t hurt my feelings.

I still use one of Jay‘s 412 IR captures quite often. At times I find that it doesn’t have enough high-end for what I’m using it for. So I apply an EQ and get the high-end that I need for the particular track or sound. I find that to be easier and actually sound a lot better in most cases than searching for a different IR.

Is that stupid?

 

[Whizzinby]

Moar realer IR’s? Ok.

 

[Orvillain]

Mixing IRs is not at issue here. It's predicting what an IR you did not acquire would be using data from neighboring positions. That is physically impossible in a general sense. If it occasionally appears to work, it's random chance, not good practice.

Riiiight, I see what you're saying. And I concur, of course. There's a similar problem in drum sampling; turning up a soft layer, or turning down a loud layer, to fill in a gap in the middle.... doesn't really work most of the time. You only get about 3-5dBrms before it sounds wrong; hence large sample sets to cover the entire dynamic range of the drum. There's a bit of clever trickery you can do to fool people, but in essence, if you didn't record it... you don't gots it.

 

[Orvillain]

OK, I’ll bite. You guys tell me if I’m completely goofy or not, and if I am, it won’t hurt my feelings.

I still use one of Jay‘s 412 IR captures quite often. At times I find that it doesn’t have enough high-end for what I’m using it for. So I apply an EQ and get the high-end that I need for the particular track or sound. I find that to be easier and actually sound a lot better in most cases than searching for a different IR.

Is that stupid?

Not goofy. Just bear in mind, you can't really add those frequencies "back" with just EQ. Once they're gone, they're gone.

Most of the time for my taste, I find a bright as hell IR; because taking away that high-end often sounds more natural than trying to fake it later if I need it.

 

[riffy]

Not goofy. Just bear in mind, you can't really add those frequencies "back" with just EQ. Once they're gone, they're gone.

Most of the time for my taste, I find a bright as hell IR; because taking away that high-end often sounds more natural than trying to fake it later if I need it.

I will try that tomorrow! Thank you!

 

[Sascha Franck]

Is that stupid?

Yes.
1) "Good is good" isn't a correct statement.
2) EQs are not there to be used. What were you thinking?

 

[Lysander]

Is that stupid?

If it sounds good, it's good. Leave the overthinking to the forums.

 

[Chocol8]

Not goofy. Just bear in mind, you can't really add those frequencies "back" with just EQ. Once they're gone, they're gone.

Most of the time for my taste, I find a bright as hell IR; because taking away that high-end often sounds more natural than trying to fake it later if I need it.

Alternatively, you could bake the EQ into the IR so that it doesn't cut as much of the high frequencies to begin with. It is a little bit of work upfront, but could be worth it if you don't have a brighter version of the IR.

 

[Sascha Franck]

Alternatively, you could bake the EQ into the IR so that it doesn't cut as much of the high frequencies to begin with. It is a little bit of work upfront, but could be worth it if you don't have a brighter version of the IR.

That's precisely what I did to get to my favourite live IRs. But more into the opposite direction, there's almost brickwall low and high cuts.

 

[jay mitchell]

I still use one of Jay‘s 412 IR captures quite often. At times I find that it doesn’t have enough high-end for what I’m using it for. So I apply an EQ and get the high-end that I need for the particular track or sound. I find that to be easier and actually sound a lot better in most cases than searching for a different IR.

Is that stupid?

Uhh, no. I often turn up the treble myself. Or (gasp!) use a high-boost shelf filter. If your only problem with an IR is its spectral tilt, you're way ahead of the game, and there's no reason to do anything but apply some EQ.

 

[jay mitchell]

Not goofy. Just bear in mind, you can't really add those frequencies "back" with just EQ. Once they're gone, they're gone.

Not gone, just lower in level than he preferred. Here's his quote: "At times I find that it doesn’t have enough high-end for what I’m using it for."

A lowpass filter with, say, a 4kHz cutoff frequency does not remove all content above 4k, it begins attenuating higher frequencies starting slightly below 4k. It's easy to recover much of the attenuated range with a filter.

The scenario in which frequencies are absent or nearly so almost always involves cancellation. If you're playing through a 4 x 12 and standing off axis, some frequencies will cancel so effectively that trying to recover them can be a waste of time.

Here's an example:

This came from an Egnater Tourmaster loaded w/Greenbacks at 15 degrees off axis horizontally and a slightly greater angle off axis vertically. The severe shelving above ca. 6.5k, could be brought more in line with the rest of the response and make the IR sound brighter without otherwise altering its character. Those frequencies aren't missing, they're just at a lower level.