Does Phase Distortion/Shift Matter in Audio? (no*)

[LIΟN]

But perhaps interesting ... on GLM i am definitely hearing the difference in this option:

Yes. You can hear the linearization below 200 Hz. What you experience is normal.

 

[restorer-john]

Other people have also purchased the books by Toole, Olive, etc.

Toole's various book/s are about as exciting as watching paint dry. You say Olive wrote a book too? Seriously, what book?

 

[j_j]

”What is almost perfect”, in a band limited sense, is the impulse response.
if one is listening to music with impulses, then it is important.

tones, and organ music, probably not so much…

Actually the impulse response fully describes what the system will do to tones. You're rather mistaken in a basic way. Way, way back there is an "FFT Workshop" at www.aes.org/sections/pnw that may help you understand how that works. There's also the "filter workshop" in the same place. I regret they are old enough that there is no recording of the presentation. Sorry.

 

[pma]

How about a nonlinear system? One impulse response would completely describe the system?

 

[j_j]

How about a nonlinear system? One impulse response would completely describe the system?

Um, in audio, we rather hope for linear systems.

Creating new frequencies, which is what nonlinear systems must do, is almost never a great thing to discover.

Ditto for time-varying systems.

 

[Cbdb2]

In those terms, literally nothing in the real world is perfect. Nature itself does not have perfect impulse response. Air pressure cannot change instantaneously, that would require infinite power. The physical world in which sounds originate and travel is power and bandwidth limited.

Also, the music we listen to doesn't exercise the full power & bandwidth of the physical world. Though sometimes it may exceed the limits of the mics used to capture it.

So the audio system doesn't actually have to be perfect. It just has to be closer to perfect than these other limiting factors.

I would say has to be closer to perfect than our hearing. Look at these examples.

Step Response: Does It Really Matter?

Below are step responses from a number of different high end / audiophile speakers: Clearly, they're all over the map and highly variable. Some look nearly "textbook" good, while others look pretty bad from a "textbook" POV. Yet many of the speakers with "bad" step responses get...

www.audiosciencereview.com

Not one has the flat top of a square wave, many have a gap between the first impulse and the second, etc, all far from "perfect" Does it matter, can we hear the difference?

 

[Holmz]

Actually the impulse response fully describes what the system will do to tones. You're rather mistaken in a basic way. Way, way back there is an "FFT Workshop" at www.aes.org/sections/pnw that may help you understand how that works. There's also the "filter workshop" in the same place. I regret they are old enough that there is no recording of the presentation. Sorry.

I am quite familiar with FFTs and windowing functions.
One could have two speakers with identical FFTs, and different impulse response curves.

Given the choice between two speakers, one with a janky impulse response, and one with an ideal impulse response… objectively I would prefer the ideal one.
It is hard for me to convince myself that sound pressure phase flips, and step responses that oscillate are true to the electrical signal that went into them.
This can be independent of whether it is something that we hear.

 

[j_j]

I am quite familiar with FFTs and windowing functions.
One could have two speakers with identical FFTs, and different impulse response curves.

Obviously you aren't familiar at all.

You may have the same POWER SPECTRUM, but you will not have the same phase spectrum in such a case.

FFT's have complex results. It's not just an amplitude or power spectrum. You will find, plainly, that two different impulse responses will have different complex spectra.

Magnitude spectrum is only half the data.

 

[Cbdb2]

The confusion for the non math guys is because most FFT software displays don't show phase.

 

[j_j]

clc
clear all
close all

len=2048;

x(1:len)=0;

x(127)=1;

xt=fft(x);

xtm=abs(xt);
xtphi=unwrap(atan2(imag(xt),real(xt)));
subplot(3,2,1)
plot(x)
axis([ 0 len -1.5 1.5]);
title('x');
subplot(3,2,3)
plot(xtm(1:len/2))
axis([ 1 len/2 -.1 1.1])
title('abs(fft(x))');
subplot(3,2,5)
plot(xtphi(1:len/2))
title('phase of xt')

yt=xt;

for ii=2:len/2
t=rand()*2*pi;
yt(ii)=yt(ii)* ( cos(t) + i * sin(t));
end
yt(len:-1len/2+2))=conj(yt(2:len/2));

y=ifft(yt);

subplot(3,2,2)
plot(y)
title('y');
subplot(3,2,4)
plot(abs(yt(1:len/2)));
axis([ 1 len/2 -.1 1.1])
title('abs(fft(y))');

ytphi=unwrap(atan2(imag(yt),real(t)));
subplot(3,2,6)
plot(ytphi(1:len/2));
title('phase of yt')


------

What you see are two signals, x and y, top row
the magnitude spectrum of x (via fft) and y (via fft) in the second row (rather similar to double precision)
the phase spectrum (third row) - yeah, not too same.

I should note this does not deal with FFT circularity, but neither do most FFT analyzers,in a very real sense.

 

[j_j]

The confusion for the non math guys is because most FFT displays don't show phase.

And that is a silly omission in some (but not all) cases. But those that lack phase info should say "magnitude of FFT", not "FFT"

 

[Cbdb2]

Given the choice between two speakers, one with a janky impulse response, and one with an ideal impulse response… objectively I would prefer the ideal one.
It is hard for me to convince myself that sound pressure phase flips, and step responses that oscillate are true to the electrical signal that went into them.
This can be independent of whether it is something that we hear.

Agreed, but as stated before, as long as you dont sacrifice other known audible characteristics, which often is the case.

 

[Cbdb2]

And that is a silly omission in some (but not all) cases. But those that lack phase info should say "magnitude of FFT", not "FFT"

Or even better, for those that think "magnitude of FFT" is the whole story, state "Phase has been omitted".

 

[j_j]

Or even better, for those that think "magnitude of FFT" is the whole story, state "Phase has been omitted".

Yeah. Not a bad idea, either.

 

[Cbdb2]

An aside. There are reverb plug ins (convolution? reverb) that you can input an in room impulse (a recorded hand clap, or the sound of film clapper) and the software emulates the full frequency response of the room reflections/ reverb. They work remarkably well. A godsend for ADR editors. You would find the clapper sound from the beginning of one of the takes and presto the ADR (usually recorded in a dead room) sounded like it was recorded on set.

 

[j_j]

https://www.aes-media.org/sections/pnw/ppt/jj/filtutv1.ppt might also be useful here.

 

[René - Acculution.com]

And the whole point of the discussion is this omitted phase.

 

[j_j]

And the whole point of the discussion is this omitted phase.

Well, omitted or not, depending on source, but that's the problem part.

 

[Holmz]

And that is a silly omission in some (but not all) cases. But those that lack phase info should say "magnitude of FFT", not "FFT"

Sometimes ^they^ say spectrum…which is perfectly fine.

The point is.. that most people use only the amplitude/magnitude
.
And the second point/.. is that the flatness of, say a pink noise spectrogram, is coveted “Uber Alles”… and the only thing many want to see.

about the only time phase is mentioned, and shown, is in a speaker’s complex load.
very seldom it is seen in measurements of the resulting pressurisation.

Obviously you aren't familiar at all.

You may have the same POWER SPECTRUM, but you will not have the same phase spectrum in such a case.

FFT's have complex results. It's not just an amplitude or power spectrum. You will find, plainly, that two different impulse responses will have different complex spectra.

Magnitude spectrum is only half the data.

The whole point of this thread topic is about phase, and whether it matters.
And I have harped that people look at the magnitude and either smile or frown.n

Agreed, but as stated before, as long as you dont sacrifice other known audible characteristics, which often is the case.

That is kinda the discussion here… how do we rank them?
do we even want to?
I think that it is important (at least to me)…
The title has in it, “does phase even matter”

Many/most stop at the flatness of the spectrum
Toole, and others, also say it is not important.

I suppose that the hairs in the ears do not know whether they are getting pushed or pulled?
But it seems like would be like seeing a tsunami come in without the water first receding.
and saying the the phase flipped tsunami is equivalent to a real tsunami.

Physically, the direction of the sound pressure seems like it should be true to the signal, if we want to call it fidelity.
But I agree that trading lots of IMD, or other distortions would be agreed to be lower fidelity for more people.

And it is hard to mathematically compare the measured sound against the electrical signal when the phase is flipped…
One sort of needs to use only the magnitude, if they want to ignore the phase.

 

[UliBru]

Take any music track of your choice.
Calculate its frequency spectrum.
Reverse the music track.
Calculate the frequency spectrum of the reversed track.
You will find that both spectra are identical.
But the music sounds different for sure