Discussion:
The Audibility of Phase Shift
(too old to reply)
Ethan Winer
2004-04-12 20:47:29 UTC
Permalink
Folks,

Often I'll hear audio engineers indict "time smear" and "phase shift" as the
reason an EQ or some other piece of gear sounds bad. I recently came across
a fabulous article written in 1981 (!) by David Clark that basically proves
phase shift per se is not audible. With David's kind permission it is now on
my web site. Here's a direct link:

www.ethanwiner.com/phase.html

Enjoy.

--Ethan
Arny Krueger
2004-04-12 21:11:38 UTC
Permalink
Post by Ethan Winer
Folks,
Often I'll hear audio engineers indict "time smear" and "phase shift"
as the reason an EQ or some other piece of gear sounds bad. I
recently came across a fabulous article written in 1981 (!) by David
Clark that basically proves phase shift per se is not audible. With
David's kind permission it is now on my web site. Here's a direct
www.ethanwiner.com/phase.html
Dave is a good friend of mine. His web site is at
http://www.dlcdesignaudio.com/ .

The phase shift audibility test circuit described in this article that Ethna
has reprinted had a another life as a component of this item:

http://www.provide.net/~djcarlst/abx_coh.htm

Unfortunately some links are broken so the schematic of the phase processor
component of the "Chamber Of Horrors" is not accessible. However, I am
intimately familiar with this device (I built part of it myself). I can
report that the schematic posted at www.ethanwiner.com/phase.html applies.
The front panel selector switch simply tapped off the signal after various
numbers of stages of all-pass network and routed it to a output circuit.

This device was demonstrated at least one AES national meeting, and also at
a number of AES section meetings. Nobody could reliably detect it, unless it
was set for totally ludicrous amounts of phase shift. A low frequency
sawtooth wave was *best* for hearing its effects. At extreme settings this
device would add a sort of a "boing" to the periodic click. The "finger
snaps" test signal you can download from
http://www.pcabx.com/technical/LR-300-3K/index.htm is a natural sound that
sounds similar and has similar effects.

Claims by people who report being able to hear the effect of a single TL074
in a signal chain should be sobered by the experience of listening to say a
half-dozen stages of the Clark/ABX phase shift circuit. Not only do you
have a bunch of TL074s cascaded, they are arguably doing something weird!

;-)
William Sommerwerck
2004-04-12 21:26:09 UTC
Permalink
Post by Ethan Winer
Often I'll hear audio engineers indict "time smear" and "phase shift"
as the reason an EQ or some other piece of gear sounds bad.
I recently came across a fabulous article written in 1981 (!) by David
Clark that basically proves phase shift per se is not audible. With
www.ethanwiner.com/phase.html
I don't see anything obviously invalid about David Clark's reasoning or
execution. However...

The second test inserts nearly constant group delay over the part of the audio
spectrum that has almost all musical fundamentals. By definition, constant group
delay is no delay at all, so this test might not be sufficiently "severe" to be
audible. Though it does seem odd that the earlier "arrival" of the top two or
three octaves _isn't_ audible.

It is well-known that phase shift above 1kHz or so is not audible for single
frequencies. But phase shift is not the same as delay or group delay.

This test certainly should be re-run with good planar speakers and electrostatic
headphones.

20 years ago, I felt the group-delay correct provided by what's-his-name's Phase
Restoration Device was audible on voice and woodwinds. My listening tests were
uncontrolled, but I did use first-generation master tapes (as that was what the
PRD was designed to correct!) and Acoustat Six electrostatic speakers.
William Sommerwerck
2004-04-12 21:26:31 UTC
Permalink
Post by Ethan Winer
Often I'll hear audio engineers indict "time smear" and "phase shift"
as the reason an EQ or some other piece of gear sounds bad.
I recently came across a fabulous article written in 1981 (!) by David
Clark that basically proves phase shift per se is not audible. With
www.ethanwiner.com/phase.html
I don't see anything obviously invalid about David Clark's reasoning or
execution. However...

The second test inserts nearly constant group delay over the part of the audio
spectrum that has almost all musical fundamentals. By definition, constant group
delay is no delay at all, so this test might not be sufficiently "severe" to be
audible. Though it does seem odd that the earlier "arrival" of the top two or
three octaves _isn't_ audible.

It is well-known that phase shift above 1kHz or so is not audible for single
frequencies. But phase shift is not the same as delay or group delay.

This test certainly should be re-run with good planar speakers and electrostatic
headphones.

20 years ago, I felt the group-delay correct provided by what's-his-name's Phase
Restoration Device was audible on voice and woodwinds. My listening tests were
uncontrolled, but I did use first-generation master tapes (as that was what the
PRD was designed to correct!) and Acoustat Six electrostatic speakers.
Arny Krueger
2004-04-12 23:04:13 UTC
Permalink
Post by William Sommerwerck
Post by Ethan Winer
Often I'll hear audio engineers indict "time smear" and "phase shift"
as the reason an EQ or some other piece of gear sounds bad.
I recently came across a fabulous article written in 1981 (!) by
David Clark that basically proves phase shift per se is not audible.
With David's kind permission it is now on my Web site. Here's a
www.ethanwiner.com/phase.html
I don't see anything obviously invalid about David Clark's reasoning
or execution. However...
The second test inserts nearly constant group delay over the part of
the audio spectrum that has almost all musical fundamentals.
Do you think that http://www.ethanwiner.com/phase.html figure 3 is an
example of constant group delay over the part of
the audio spectrum that has almost all musical fundamentals? For example,
there's an approximate 30% drop in group delay between 200 Hz and 1 KHz.

http://www.mmedia.is/~quinton/ch7.html
says that "The working range of fundamental frequencies, the range of tones
produced on musical instruments, is then approximately 27 to 4200 cycles per
second." This is consistent with other articles on the topic.
http://www.ethanwiner.com/phase.html figure 3 shows an approximate 80%
drop in group delay over this range.
Post by William Sommerwerck
This test certainly should be re-run with good planar speakers and
electrostatic headphones.
Clark has had large Magnepalanar speakers in his primary listening rooms
since some time in the late 1980s. He has done numerous demonstrations of
the audibility of phase shift since then. I believe that Stax electrostats
were used in some of his AES group presentations.

Anybody who wants to can repeat any of the tests that they can freely
download from http://www.pcabx.com/technical/LR-300-3K/index.htm and
http://www.pcabx.com/technical/polarity/index.htm with whatever reproducers
they choose to use. True for the entire site, of course.
William Sommerwerck
2004-04-13 01:14:52 UTC
Permalink
Post by Arny Krueger
Post by William Sommerwerck
The second test inserts nearly constant group delay over the part of
the audio spectrum that has almost all musical fundamentals.
Do you think that http://www.ethanwiner.com/phase.html figure 3 is an
example of constant group delay over the part of the audio spectrum
that has almost all musical fundamentals? For example, there's an
approximate 30% drop in group delay between 200 Hz and 1 KHz.
I didn't say constant. I said nearly constant. A 30% drop is not huge,
especially as it's a gradual change. I'd like to see a more abrupt, extreme
change within the 100 to 3k band.

Simply out of curiosity -- did Mr. Clark look at any of the musical waveforms on
a 'scope?
Post by Arny Krueger
Post by William Sommerwerck
This test certainly should be re-run with good planar speakers
and electrostatic headphones.
Clark has had large Magnepalanar speakers in his primary
listening rooms since some time in the late 1980s.
Large Maggies of that era are absolutely NOT good planar speakers.
Post by Arny Krueger
He has done numerous demonstrations of the audibility of phase shift
since then. I believe that Stax electrostats were used in some of his
AES group presentations.
If he got the same results (which I assume you're implying), I'd be highly
inclined to believe those.
Arny Krueger
2004-04-13 01:41:44 UTC
Permalink
This post might be inappropriate. Click to display it.
Mike Rivers
2004-04-13 00:46:59 UTC
Permalink
Post by Ethan Winer
Often I'll hear audio engineers indict "time smear" and "phase shift" as the
reason an EQ or some other piece of gear sounds bad. I recently came across
a fabulous article written in 1981 (!) by David Clark that basically proves
phase shift per se is not audible.
The reason for this belief is that people are (yet again) mis-using the
term "phase." What makes EQ sound bad is group delay - when a group
of frequencies (which may be a single instrument playing a note with
overtones) which have a given set of phase relationships go through a
network which changes those phase relationships. The squiggles change,
so the timbre changes.



--
I'm really Mike Rivers (***@d-and-d.com)
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me here: double-m-eleven-double-zero at yahoo
William Sommerwerck
2004-04-13 01:15:52 UTC
Permalink
Post by Mike Rivers
The reason for this belief is that people are (yet again) mis-using the
term "phase." What makes EQ sound bad is group delay - when a group
of frequencies (which may be a single instrument playing a note with
overtones) which have a given set of phase relationships go through a
network which changes those phase relationships. The squiggles change,
so the timbre changes.
But Clark's test (sort of) suggest that relatively large amounts of non-linear
group delay aren't audible. (See my other posting.)
Scott Dorsey
2004-04-13 02:12:05 UTC
Permalink
Post by Mike Rivers
Post by Ethan Winer
Often I'll hear audio engineers indict "time smear" and "phase shift" as the
reason an EQ or some other piece of gear sounds bad. I recently came across
a fabulous article written in 1981 (!) by David Clark that basically proves
phase shift per se is not audible.
The reason for this belief is that people are (yet again) mis-using the
term "phase." What makes EQ sound bad is group delay - when a group
of frequencies (which may be a single instrument playing a note with
overtones) which have a given set of phase relationships go through a
network which changes those phase relationships. The squiggles change,
so the timbre changes.
Group delay through typical speakers is so bad, though, that even severe
midrange group delay may be inaudible because it's masked by the crossover
effects in the speaker. The fact that multiway speakers sound any good
at all points out how difficult it is to hear group delay effects.

On the other hand, constant time delay (zero group delay, constant time
delay for all frequencies) is what tape machines are built for. You put
a waveform in, and phase shift it thirty years.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
James Cunningham
2004-04-13 02:35:47 UTC
Permalink
Post by Scott Dorsey
Post by Mike Rivers
Post by Ethan Winer
Often I'll hear audio engineers indict "time smear" and "phase shift" as the
reason an EQ or some other piece of gear sounds bad. I recently came across
a fabulous article written in 1981 (!) by David Clark that basically proves
phase shift per se is not audible.
The reason for this belief is that people are (yet again) mis-using the
term "phase." What makes EQ sound bad is group delay - when a group
of frequencies (which may be a single instrument playing a note with
overtones) which have a given set of phase relationships go through a
network which changes those phase relationships. The squiggles change,
so the timbre changes.
Group delay through typical speakers is so bad, though, that even severe
midrange group delay may be inaudible because it's masked by the crossover
effects in the speaker. The fact that multiway speakers sound any good
at all points out how difficult it is to hear group delay effects.
On the other hand, constant time delay (zero group delay, constant time
delay for all frequencies) is what tape machines are built for. You put
a waveform in, and phase shift it thirty years.
--scott
Is that why I've seen 1KHz square waves that look nearly perfect on a scope?
Jim C.
Bob Cain
2004-04-13 01:02:24 UTC
Permalink
Post by Ethan Winer
Folks,
Often I'll hear audio engineers indict "time smear" and "phase shift" as the
reason an EQ or some other piece of gear sounds bad. I recently came across
a fabulous article written in 1981 (!) by David Clark that basically proves
phase shift per se is not audible. With David's kind permission it is now on
www.ethanwiner.com/phase.html
I would be interested in the testing of a system where the
upper frequencies were delayed with respect to the lower
rather than the other way around. I believe, based on
informal listening tests with linear verses minimum phase
filters that this would be more detectable.


Bob
--
"Things should be described as simply as possible, but no
simpler."

A. Einstein
Arny Krueger
2004-04-13 01:48:09 UTC
Permalink
Post by Bob Cain
I would be interested in the testing of a system where the
upper frequencies were delayed with respect to the lower
rather than the other way around. I believe, based on
informal listening tests with linear verses minimum phase
filters that this would be more detectable.
You've got Audition, right? Don't let me hold you back!

Just split the bands with the FFT filter. Time-shift them by removing some
samples from one of them. Mix/paste to put Humpty back together again.

For practical purposes, Auditiion's FFT filter has no phase shift. People
have even asked Adobe to make phase shift an option...

You can check this method out by taking some music apart and putting it back
together and comparing the final results to the starting point with PCABX.
David Collins
2004-04-14 07:08:05 UTC
Permalink
Post by Bob Cain
I would be interested in the testing of a system where the
upper frequencies were delayed with respect to the lower
rather than the other way around. I believe, based on
informal listening tests with linear verses minimum phase
filters that this would be more detectable.
Then listen to any CD mastered before, say, 1987 as they will have
like 1ms delay at Nyquist from the multi-pole analog LPF's.

If your ear was really bothered by high-frequency group delay, you
wouldn't be able to tell what song it was...

This didn't stop me from implementing allpass filters on early PCM
systems, but I'm not sure if it did anything. maybe another half dB
less overshoot. At 1k...

DC
--
Dave Collins Entropy just isn't what it used to be!


www.collinsaudio.com
Bob Olhsson
2004-04-13 16:30:43 UTC
Permalink
...article written in 1981 (!) by David Clark that basically proves
phase shift per se is not audible.
Dave Clark wrote: "Listeners were simply asked to perform a task or to hear
a difference. I feel that studies of this type are more relevant to
knowledge of sound reproduction than clinical tests or opinions derived from
"open" evaluations."

It's a well established fact that blind listening tests done without any
training of the subjects using a graphic demonstration of the artifacts
being studied will virtually always produce random results. Non-"clinical"
double blind tests are pseudoscience that prove nothing.

I understand it's been established for a long time using clinical tests by
organizations such as Bell Labs that phase shift is very much audible to at
least some people.
--
Bob Olhsson Audio Mastery, Nashville TN
Mastering, Audio for Picture, Mix Evaluation and Quality Control
Over 40 years making people sound better than they ever imagined!
615.385.8051 http://www.hyperback.com
Ethan Winer
2004-04-14 14:07:25 UTC
Permalink
Bob,
Post by Bob Olhsson
I understand it's been established for a long time using clinical tests by
organizations such as Bell Labs that phase shift is very much audible to at
least some people. <

I think it's important to distinguish between normal amounts of phase shift
that occur in nature - and in all analog EQ circuits - and the huge amounts
in this contrived example. Even David Clark acknowledges in his article that
the extreme shift from 15 all-pass stages was clearly audible on certain
percussive sounds.

I did some experimenting with this about 20 years ago, and I wasn't able to
hear the shift from three all-pass stages. I may not have the best ears in
the biz, but I'm far from an untrained listener. Even if really good
listeners can discern a small change in timbre, phase shift is not nearly
the gremlin it's made out to be.

I often hear people blame phase shift for causing a "phasey" sound when they
boost an EQ. But phase shift won't sound "phasey" like a flanger effect.
Rather, they're hearing comb filtering that's already present - probably
from reflections off a nearby wall - and boosting the treble simply brings
that out in the track.

--Ethan
Scott Dorsey
2004-04-14 14:24:40 UTC
Permalink
Post by Bob Olhsson
Post by Bob Olhsson
I understand it's been established for a long time using clinical tests by
organizations such as Bell Labs that phase shift is very much audible to at
least some people. <
I think it's important to distinguish between normal amounts of phase shift
that occur in nature - and in all analog EQ circuits - and the huge amounts
in this contrived example. Even David Clark acknowledges in his article that
the extreme shift from 15 all-pass stages was clearly audible on certain
percussive sounds.
How does that compare with the extreme shift from typical speakers? I see
a lot of speakers with more the 360 degrees of phase shift across the
spectrum. I would be very surprised if this is not audible.
Post by Bob Olhsson
I did some experimenting with this about 20 years ago, and I wasn't able to
hear the shift from three all-pass stages. I may not have the best ears in
the biz, but I'm far from an untrained listener. Even if really good
listeners can discern a small change in timbre, phase shift is not nearly
the gremlin it's made out to be.
I did something like this too, and I also couldn't really hear it, but by
the same token, I sure could hear artifacts from early brickwall filters on
A/D systems. I won't absolutely positively say that what I heard was due to
group delay, but it would seem reasonable.
Post by Bob Olhsson
I often hear people blame phase shift for causing a "phasey" sound when they
boost an EQ. But phase shift won't sound "phasey" like a flanger effect.
Rather, they're hearing comb filtering that's already present - probably
from reflections off a nearby wall - and boosting the treble simply brings
that out in the track.
Not to mention all of the other artifacts that come along with typical
EQ networks, like slew-limiting and all kinds of interesting transient
distortion. If you look at the Bode plot on an OPA604, it has a loop in
it for God's sake. You can imagine how stable THAT must be in a gyrator
circuit.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Arny Krueger
2004-04-14 14:32:44 UTC
Permalink
Post by Scott Dorsey
Post by Bob Olhsson
Post by Bob Olhsson
I understand it's been established for a long time using clinical tests by
organizations such as Bell Labs that phase shift is very much
audible to at least some people. <
I think it's important to distinguish between normal amounts of
phase shift that occur in nature - and in all analog EQ circuits -
and the huge amounts in this contrived example. Even David Clark
acknowledges in his article that the extreme shift from 15 all-pass
stages was clearly audible on certain percussive sounds.
How does that compare with the extreme shift from typical speakers?
I see a lot of speakers with more the 360 degrees of phase shift
across the spectrum. I would be very surprised if this is not
audible.
I think the consensus of careful listener who have tried the DBTs posted at
http://www.pcabx.com/technical/LR-300-3K/index.htm , which is basically a
simulation of an acoustical Linkwitz-Riley speaker system (360 degree phase
shift and all), is that many people can hear slight differences with the
finger snaps. Anybody who doubts should listen for themselves. The price is
*right*!

So don't be surprized! I think we can say that it's audible.

However, compared to the other audible artifacts that every speaker has...
Linkwitz-Riley was never presented as a panacea, just a beneficial
technology that addresses far more significant problems than the one it
creates.
Ethan Winer
2004-04-15 14:54:32 UTC
Permalink
Scott,
I see a lot of speakers with more the 360 degrees of phase shift across
the spectrum. I would be very surprised if this is not audible. <

This is exactly what tests like these are designed to determine. Versus
assuming that an audible effect must be phase shift. As we all know, there
are all kinds of horrible things going on in every loudspeaker. To my way of
thinking, phase shift is near the bottom of the list. Way after distortion
and cone resonances.
I sure could hear artifacts from early brickwall filters on A/D systems. <
How do you know it was the phase shift, as opposed to any of a dozen other
possibilities? How do you even know it was the filters you heard?

--Ethan
Scott Dorsey
2004-04-15 15:07:23 UTC
Permalink
Post by Ethan Winer
I see a lot of speakers with more the 360 degrees of phase shift across
the spectrum. I would be very surprised if this is not audible. <
This is exactly what tests like these are designed to determine. Versus
assuming that an audible effect must be phase shift. As we all know, there
are all kinds of horrible things going on in every loudspeaker. To my way of
thinking, phase shift is near the bottom of the list. Way after distortion
and cone resonances.
Exactly. If phase shift was the main issue, speakers would sure sound a
lot better than they do. But there are a lot of speakers on which it is
clear that something is just wrong in the crossover region. That something
might just be destructive interference issues off-axis, but it might also
be total group delay on-axis.

The thing is that the amount of phase shift in a typical speaker system is
just enormous. It's incredibly greater than anything you'll find in typical
electronics. This is unfortunately confounded by the fact that typical
speaker systems also have a lot more distortion than you'll ever find in
typical electronics too, and it can be hard to sort the two out.

But none of the tests I have seen have been modelling anywhere near the
sort of enormous group delay that you see on speakers. We're not talking
about three or four all-pass networks here... we're talking hundreds of
times worse trouble.
Post by Ethan Winer
I sure could hear artifacts from early brickwall filters on A/D systems. <
How do you know it was the phase shift, as opposed to any of a dozen other
possibilities? How do you even know it was the filters you heard?
I know it was the filters, because upgrading the filters (say, with the
Apogee kits on the PCM-1630) improved things. But it may not have been
phase shift, it's true. Again, some of these early anti-aliasing and
reconstruction filters just had such gargantuan group delay numbers that
transients were visibly mangled on a 3" scope.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
William Sommerwerck
2004-04-15 17:31:46 UTC
Permalink
Post by Scott Dorsey
If phase shift was the main issue, speakers would sure sound a
lot better than they do. But there are a lot of speakers on which
it is clear that something is just wrong in the crossover region.
That something might just be destructive interference issues
off-axis, but it might also be total group delay on-axis.
20 years ago, when I worked for Rupert Neve, I heard a pair of multi-driver
custom monitors designed by a refugee from the aerospace industry. They were
installed at Atlanta's biggest studio (I forget the name), mounted about 8' off
the floor in the wall above the window into the performance area.

They were so poorly designed that when you played noise through them, you could
instantly localize each individual driver as a separate sound source, even
though the systems were a good 10' from your ears! On the other hand, I've heard
lots of audiophile speakers, from more than one company, that sound like a
single driver even when you poke your head right in front of them.

I don't want to speculate on why these speakers were so bad. I'm just presenting
the observation.
Post by Scott Dorsey
The thing is that the amount of phase shift in a typical speaker
system is just enormous. It's incredibly greater than anything
you'll find in typical electronics. This is unfortunately confounded
by the fact that typical speaker systems also have a lot more
distortion than you'll ever find in typical electronics too, and it
can be hard to sort the two out.
But none of the tests I have seen have been modelling anywhere
near the sort of enormous group delay that you see on speakers.
We're not talking about three or four all-pass networks here...
we're talking hundreds of times worse trouble.
I hate to sound as if I'm agreeing with Arny, but...

If you believe "High Performance Loudspeakers" (which, interestingly, largely
ignores planar systems), well-designed drivers are usually minimum-phase. Even
if they're combined with a network that is flat only in the amplitude domain
(thus producing a non-minimum-phase system with large amounts of group delay),
the delay is not "enormous." Whether it's audible is another matter.
Arny Krueger
2004-04-15 18:19:55 UTC
Permalink
Post by Scott Dorsey
The thing is that the amount of phase shift in a typical speaker
system is just enormous. It's incredibly greater than anything
you'll find in typical electronics.
Agreed. You don't find 360 degree phase shifts at 300 and 3000 Hz in typical
electronics, unless that "typical electronics" is a 24 dB/octave
Linkwitz-Riley crossover or something that incorporates that kind of a
function.
Post by Scott Dorsey
This is unfortunately confounded
by the fact that typical speaker systems also have a lot more
distortion than you'll ever find in typical electronics too, and it
can be hard to sort the two out.
I put that under the slings and arrows of technical analysis and subjective
analysis.
Post by Scott Dorsey
But none of the tests I have seen have been modeling anywhere near
the sort of enormous group delay that you see on speakers.
I wouldn't be too sure of that. As Bill points out, a speaker driver not
operating in breakup mode can be counted on to be a minimum phase device if
you are measuring at or compensating for the acoustic center.

The "Acoustical Linkwitz-Riley Speaker System, 360 degree phase shifts and
all" can IME be a real-world device. We also had Dunlavy speakers that had
both reasonably smooth frequency response and fair transient response with a
variety of tests, at least on-axis.
Post by Scott Dorsey
We're not talking about three or four all-pass networks here... we're
talking
Post by Scott Dorsey
hundreds of times worse trouble.
That would be LOTS. Even good speakers are kinda scary to measure, but as a
rule the better ones aren't THAT bad. Not 100 times worse than a Dunlavy and
not 100 times worse than an Acoustical Linkwitz-Riley Speaker System.

Now if you say twice as bad or something like that, no argument. However, a
speaker with a 360 degree phase shift at 300 degrees involves a equivalent
path length difference of about 3 feet. 100 times that is 300 feet. I
don't think we have many real-world speakers with as much phase shift as a
football field.

This "100 times" was like a figure of speech, right?

;-)
Scott Dorsey
2004-04-15 21:29:45 UTC
Permalink
Post by Arny Krueger
Now if you say twice as bad or something like that, no argument. However, a
speaker with a 360 degree phase shift at 300 degrees involves a equivalent
path length difference of about 3 feet. 100 times that is 300 feet. I
don't think we have many real-world speakers with as much phase shift as a
football field.
Well, I got a plot of a speaker around here somewhere that goes around
the clock two and a half times. That's nine hundred degrees. Admittedly
this is a degenerate case.
Post by Arny Krueger
This "100 times" was like a figure of speech, right?
Well, nine degrees seems way more than acceptable for an amplifier.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Arny Krueger
2004-04-15 21:40:11 UTC
Permalink
Post by Scott Dorsey
Post by Arny Krueger
Now if you say twice as bad or something like that, no argument.
However, a speaker with a 360 degree phase shift at 300 degrees
involves a equivalent path length difference of about 3 feet. 100
times that is 300 feet. I don't think we have many real-world
speakers with as much phase shift as a football field.
Well, I got a plot of a speaker around here somewhere that goes around
the clock two and a half times. That's nine hundred degrees.
Admittedly this is a degenerate case.
Post by Arny Krueger
This "100 times" was like a figure of speech, right?
Well, nine degrees seems way more than acceptable for an amplifier.
Oh, the "100 times" is based on a comparison between a really bad speaker
and an amp.

Got it!
Chris Hornbeck
2004-04-15 23:12:54 UTC
Permalink
Post by Arny Krueger
As Bill points out, a speaker driver not
operating in breakup mode can be counted on to be a minimum phase device if
you are measuring at or compensating for the acoustic center.
One caveat though; the "acoustic center" is very slippery. It's
only kissing cousins with either the motor or the innermost
radiating surface, and attempts to measure it seem to cause it
to run and hide. The models fail badly here.
Post by Arny Krueger
The "Acoustical Linkwitz-Riley Speaker System, 360 degree phase shifts and
all" can IME be a real-world device.
The other caveat is WRT an axis of measurement. Linkwitz crossovers
sum to unity on axis at infinity, or with no separation between
drivers. Real-world implementations commonly have about a whole
wavelength separation, and use and measurements at tens of
wavelengths.

FWIW, D'Appolito's geometric solution allows Butterworth filters,
which have the group delay characteristics in an n-order crossover
of an n+1 order L-R crossover, and flat summed whole-room response
even with non-zero driver separation.

Chris Hornbeck
Chris Hornbeck
2004-04-15 23:29:20 UTC
Permalink
On Thu, 15 Apr 2004 23:12:54 GMT, Chris Hornbeck
Post by Chris Hornbeck
FWIW, D'Appolito's geometric solution allows Butterworth filters,
which have the group delay characteristics in an n-order crossover
of an n+1 order L-R crossover,
I can't get the simplest thing right. Should be "in an n+1 order
crossover of an n order L-R crossover"

Chris Hornbeck
Ethan Winer
2004-04-16 13:36:31 UTC
Permalink
Scott,
there are a lot of speakers on which it is clear that something is just
wrong in the crossover region. That something might just be destructive
interference issues off-axis, but it might also be total group delay
on-axis. <

This is the main point Dave makes in the first part of his article. Phase
shift from the crossover, or from the drivers' physical alignment, can
clearly affect the sound. But the real culprit is a skewed *frequency
response* caused when the drivers' outputs combine in the air. Not the phase
shift itself.

--Ethan
Scott Dorsey
2004-04-16 13:47:22 UTC
Permalink
Post by Ethan Winer
Scott,
there are a lot of speakers on which it is clear that something is just
wrong in the crossover region. That something might just be destructive
interference issues off-axis, but it might also be total group delay
on-axis. <
This is the main point Dave makes in the first part of his article. Phase
shift from the crossover, or from the drivers' physical alignment, can
clearly affect the sound. But the real culprit is a skewed *frequency
response* caused when the drivers' outputs combine in the air. Not the phase
shift itself.
That is a major issue. The question is whether the phase shift on-axis is
also audible, and that's a question that is hard to answer because any
effects are going to be masked by the (much more audible) off-axis interference
issues. So I'm not buying that the on-axis group delay isn't also a
possible second problem, just that it's hard to tell if it is or not.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Ethan Winer
2004-04-16 20:34:00 UTC
Permalink
Scott,
I'm not buying that the on-axis group delay isn't also a possible second
problem, just that it's hard to tell if it is or not. <

Okay, I know you well enough not to argue with you!

:->)

--Ethan
Scott Dorsey
2004-04-17 00:04:46 UTC
Permalink
Post by Ethan Winer
I'm not buying that the on-axis group delay isn't also a possible second
problem, just that it's hard to tell if it is or not. <
Okay, I know you well enough not to argue with you!
No, no, it's a good argument. And if you can propose a way to really
tell if this stuff is audible, I'll help you implement it. I don't
really know and I'm not sure anyone really does, and I'd like to find
out.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Ethan Winer
2004-04-17 13:00:20 UTC
Permalink
Scott,
if you can propose a way to really tell if this stuff is audible, I'll
help you implement it. I don't really know and I'm not sure anyone really
does, and I'd like to find out. <

Well, I thought Dave's test was pretty convincing, and my own test (years
ago) showed me that phase shift is not audible. Again, with *extreme*
amounts, on certain types of program material, it can be audible. My reason
for raising this now is constantly seeing people blame phase shift for
things that are clearly frequency response errors or other problems.

As for implementing further tests, I don't consider myself qualified. This
is more a job for Arny and the other smart guys here - like you.

--Ethan
Scott Dorsey
2004-04-17 14:03:29 UTC
Permalink
Post by Ethan Winer
if you can propose a way to really tell if this stuff is audible, I'll
help you implement it. I don't really know and I'm not sure anyone really
does, and I'd like to find out. <
Well, I thought Dave's test was pretty convincing, and my own test (years
ago) showed me that phase shift is not audible. Again, with *extreme*
amounts, on certain types of program material, it can be audible. My reason
for raising this now is constantly seeing people blame phase shift for
things that are clearly frequency response errors or other problems.
The test is a good start.

I can hear a difference, though, between a digital 4 KHZ low-pass filter,
second order only, FIR and IIR implementations. And in theory, if we aren't
dealing with any rounding issues, the only difference between these two are
in the phase response.

A test, though, is going to have to involve (presumably digital) filters
with controllable phase response, and most importantly a playback system
that has as flat as possible phase response. If the group delay of the
playback system is swamping that of the test filter, you're not going to
see significant effects.

How do we get a playback system with that flat phase response in the first
place? And, can we conduct tests on a limited spectrum range, thereby
reducing the range over which the playback system needs flat phase response?

I agree that people do blame group delay for things that probably don't
have anything to do with it, but I just don't want to throw out the baby
with the bathwater. I want to know where "extreme amounts" begin and end.
And I want to know how distortion effects can mask group delay effects.
Post by Ethan Winer
As for implementing further tests, I don't consider myself qualified. This
is more a job for Arny and the other smart guys here - like you.
I'm too busy fixing goddamn Volumaxes, attending meaningless conferences,
and posting to Usenet to get any actual work done.
--scott

Oh yeah, I do have an actual group delay measurement of an AM broadcast
station before and after fixing the antenna system. The plot is just
scary. That's another thing I have been doing in place of real work.
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
unknown
2004-04-18 01:06:00 UTC
Permalink
About Some experiment with time...

What David Clarke humbly tried to prove is only that phase audibility is not a first order priority and non-linear magnitude frequency response is much more relevant.
Maybe, when all what affects linearity will be almost perfectly controlled and all types of distortion will be reduced to a very low level, then waveform distortion will jump up in the priority list.


Luke




----------
Sent via SPRACI - http://www.spraci.net/ - Parties,Raves,Clubs,Festivals
Scott Dorsey
2004-04-18 01:29:30 UTC
Permalink
Post by unknown
What David Clarke humbly tried to prove is only that phase audibility is not a first order priority and non-linear magnitude frequency response is much more relevant.
Yes, that's clear and doesn't really need proving at this point. I want
to know at what point group delay DOES become audible, though.

In part this is because I often see systems in which group delay is so
substantial that it might be an issue. Those systems include loudspeakers
and broadcast transmitters. Yes, these are degenerate cases, but they are
degenerate cases we encounter all too often.
Post by unknown
Maybe, when all what affects linearity will be almost perfectly controlled and all types of distortion will be reduced to a very low level, then waveform distortion will jump up in the priority list.
If we take loudspeakers completely out of the equation, this is reasonable
to expect. If we leave loudspeakers in the equation, it's not going to
happen for a long time.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
unknown
2004-04-18 02:33:18 UTC
Permalink
Post by Scott Dorsey
to know at what point group delay DOES become audible, though.
Blauert and Laws Criterias shows about 1mS at 2KHz. Abobe this frequency it seems more GD is needed for detection. Below,it's even more. Hey! That is equivalent to an 2KHz-8th-order LR crossover GD distortion.
Group Delay Distortion in electro-acoustic systems, J.Blauert, P.Laws, JAS vol.63, pp1478-1483

If we leave loudspeakers in the equation, it's not going to
Post by Scott Dorsey
happen for a long time
You're right! A lot of water will flow under the bridge since then!

To make it short, phase by itself is not audible. When its value makes it relevant to affect FR during some sort of summing as in loudspeakers then FR becomes audible. Same thing when a filter a sufficient order is employed. Then ringing is what becomes audible. And even more when using worse than a Bessel filter.

Luke
----------
Sent via SPRACI - http://www.spraci.net/ - Parties,Raves,Clubs,Festivals
unknown
2004-04-18 02:33:13 UTC
Permalink
Post by Scott Dorsey
to know at what point group delay DOES become audible, though.
Blauert and Laws Criterias shows about 1mS at 2KHz. Abobe this frequency it seems more GD is needed for detection. Below,it's even more. Hey! That is equivalent to an 2KHz-8th-order LR crossover GD distortion.
Group Delay Distortion in electro-acoustic systems, J.Blauert, P.Laws, JAS vol.63, pp1478-1483

If we leave loudspeakers in the equation, it's not going to
Post by Scott Dorsey
happen for a long time
You're right! A lot of water will flow under the bridge since then!

To make it short, phase by itself is not audible. When its value makes it relevant to affect FR during some sort of summing as in loudspeakers then FR becomes audible. Same thing when a filter a sufficient order is employed. Then ringing is what becomes audible. And even more when using worse than a Bessel filter.

Luke
----------
Sent via SPRACI - http://www.spraci.net/ - Parties,Raves,Clubs,Festivals
Ethan Winer
2004-04-18 13:23:41 UTC
Permalink
Luke,
Post by unknown
What David Clarke humbly tried to prove is only that phase audibility is
not a first order priority and non-linear magnitude frequency response is
much more relevant. <

That's a beautiful one-sentence summary, and I agree completely.

It really doesn't matter to me whether extreme amounts of phase shift are
audible. It's not much of a problem in the overall scheme of things. The
same could be said for ultra-high sample rates and bit depths, the value of
passive summing, and a dozen other such things that are discussed far more
than deserved.

Too many people who have *no idea* how audio really works blame phase shift
because they continually read magazine and web articles saying phase shift
is a problem. Interviews with their favorite artists and producers parrot
the same words. I've been following this nonsense for 25+ years now. I first
noticed "phase as mystique" in the ads for the original Aphex Aural Exciter,
which was total BS and lied saying it's based on phase shift. It's a
glorified fuzz box, fer cryin' out loud, but I guess that doesn't have the
right marketing pizzazz.

It's been all downhill since then.

--Ethan
Scott Dorsey
2004-04-18 13:46:16 UTC
Permalink
Post by Ethan Winer
I've been following this nonsense for 25+ years now. I first
noticed "phase as mystique" in the ads for the original Aphex Aural Exciter,
which was total BS and lied saying it's based on phase shift. It's a
glorified fuzz box, fer cryin' out loud, but I guess that doesn't have the
right marketing pizzazz.
The BBE box uses the same marketing, and basically does much of the same
thing inside that the Aphex does (although with a very different spectrum).
It's very interesting to compare what the manual to the BBE says that it
does with what it actually does.

Little Labs makes a little all-pass network that you can use to screw around
with signal phase, to deal with signals from delayed sources in multitrack
mixes. Their marketing says that it sounds quite transparent even with
large amounts of group delay dialed in. So it's nice to see it go both ways.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Ethan Winer
2004-04-19 13:39:31 UTC
Permalink
Scott,
The BBE box uses the same marketing <
Yes, and I thought about mentioning BBE but was too lazy. Yet more marketing
BS.
Little Labs makes a little all-pass network <
Yes, and that's a great way for people to do the same experiment I tried
20-odd years ago. Except the In-Between Phase box goes 0-360 only. But it
still should be enough to convince the "phase believers" it's a non-issue.

Thanks to everyone else for all the other comments too. I'm glad to see this
out in the open.

--Ethan
Arny Krueger
2004-04-19 15:42:28 UTC
Permalink
Post by Ethan Winer
Thanks to everyone else for all the other comments too. I'm glad to
see this out in the open.
While we're talking about phase shift, why not talk about a sitaution where
relatively small amounts of in-line phase shift seems to be audible?

I'm talking about the pervasive high-pass filtering that is all over the
audio production process.

Last week I did some PCABXing of a piece of music I made by stringing
together 5 copies of the same kick drum sample.

I guess that if I was going to do something formal with this concept I'd
want a kick drum sample with a pedigree, but this one seemed to sound and
look about right:

http://www.users.bigpond.com/prodigalson/waves.htm

http://www.users.bigpond.com/prodigalson/Sounds/kicks_01.zip

and I used tight_kick_18.wav from that zip.

I then PCABXed the original versus the same thing only high-pass filtered. I
started out with a high pass filter from Audition that was sixth order
Butterworth, to get an order of magnitude idea of how obvious the changes
are. Later on I did some tests with lower order filters, particularly third.
I think its pretty hard to set up a playback chain that can be characterized
as something as simple as even a third order high pass filter. Most are
going to be higher order, and therefore likely much easier to hear.

I started out with a corner frequency that was 50 Hz, and worked down until
I couldn't get perfect scores very easily. I'm not talking about a tough
listening test, I'm talking about something that is pretty obvious.
Probability more obvious to some others than me. That happened with a
corner frequency around 2 Hz. That seems to set a pretty high bar for
sonically perfect reproduction of what seems like a relatively easy sound,
no?

Looking at the before and after waves, it seems to me that phase shift is
the predominant effect. At the lower corner frequencies, there isn't that
much visible distortion of the wave, and the over-all amplitude does not
change that much.

Physiologically speaking, there's no reason I know of to presume that phase
shift at 20 Hz and below wouldn't be in some sense audible. The seems to
have the hardware to do the job, unlike the situation at 500-1 KHz and
above.
William Sommerwerck
2004-04-19 18:07:07 UTC
Permalink
Post by Arny Krueger
Looking at the before and after waves, it seems to me that
phase shift is the predominant effect. At the lower corner
frequencies, there isn't that much visible distortion of the
wave, and the over-all amplitude does not change that much.
Physiologically speaking, there's no reason I know of to
presume that phase shift at 20 Hz and below wouldn't be in
some sense audible. The seems to have the hardware to do
the job, unlike the situation at 500-1 KHz and above.
This is not a new observation. Martin Colloms (in "High-Performance
Loudspeakers") claimed that non-linear group delay was audible at low
frequencies. Of course, since the ear _can_ hear static phase differences below
1kHz, it has not been a controversial view.
Arny Krueger
2004-04-19 18:22:55 UTC
Permalink
Post by William Sommerwerck
Post by Arny Krueger
Looking at the before and after waves, it seems to me that
phase shift is the predominant effect. At the lower corner
frequencies, there isn't that much visible distortion of the
wave, and the over-all amplitude does not change that much.
Physiologically speaking, there's no reason I know of to
presume that phase shift at 20 Hz and below wouldn't be in
some sense audible. The ear seems to have the hardware to do
the job, unlike the situation at 500-1 KHz and above.
This is not a new observation. Martin Colloms (in "High-Performance
Loudspeakers") claimed that non-linear group delay was audible at low
frequencies.
I'm quite sure that it's not a new observation. Like many things, the charm
if there is any, is in

(1) Proven reliability of the observation. Claim versus fact.

(2) Quantification of the thing that is observed. Specifically, what are
"low frequencies"?

(3) Relevance to common situations. If low frequency phase shift is a
detriment to sound quality, how often does it happen?

(4) Possibility of addressing the situation and obtaining a positive
outcome.

It looks to me like establishing relevant facts is reasonably easy. When I
say "low frequencies" I'm talking 10 Hz and below, maybe as low as 1-2 Hz.
It appears that this form of coloration is pretty common in recorded and
reproduced music. The more pervasively that digitally-based production
methods are used, and reasonable hygiene is followed, the more likely there
won't be a problem from this source.
Post by William Sommerwerck
Of course, since the ear _can_ hear static phase
differences below 1kHz, it has not been a controversial view.
How much of that is dependent on the use of two ears?

Since the effect I'm talking about involves a signal where the two channels
started out identical, and were altered equally, it does not seem to depend
much on the use of two ears.
Fill X
2004-04-19 15:46:29 UTC
Permalink
This post might be inappropriate. Click to display it.
Scott Dorsey
2004-04-19 15:55:07 UTC
Permalink
Post by Fill X
Post by Ethan Winer
xcept the In-Between Phase box goes 0-360 only. But it
still should be enough to convince the "phase believers" it's a non-issue.
sorry ethan, what are you meaning here about "phase believers" and the little
labs box. Seems like a sound idea to bein the intended application.
Oh what can it mean
For a phase believer
And a homecoming queen....
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Ethan Winer
2004-04-20 12:51:35 UTC
Permalink
Philip,
what are you meaning here about "phase believers" and the little labs box.
Seems like a sound idea to bein the intended application. <

I did not at all mean to imply the LittleLabs box is not useful! The point
is you can *optionally* use that as a test device on a single channel to see
if you can hear phase shift.

--Ethan
William Sommerwerck
2004-04-18 14:03:36 UTC
Permalink
It might be a good idea to point out that "the audibility of phase shift" is not
really what we're talking about.

What we should call the subject is "the audibility of non-linear group delay."
unknown
2004-04-18 15:34:05 UTC
Permalink
Ethan

You're right. "Phase" is a real buzz-word in the world of audio.
A kind of inquisition tribunal for which all audible evils are adressed. "Oh! I hear something bad...HaaaH! It must be a phase problem." I ever heard a guy who was hearing a phase problem in an amplifier at 500Hz. He claimed few degrees. This one is special. I also know a friend who bought a fourth power amplifier for a Linn tri-amped system. He said this is a spare one in case of the company would modify future version of it. So he would keep the same phase response in his system. Phase is a real buzzword and it works.

All this is known by the industry and is reflected in marketing.

Luke






----------
Sent via SPRACI - http://www.spraci.net/ - Parties,Raves,Clubs,Festivals
unknown
2004-04-18 15:42:13 UTC
Permalink
Sorry...bad publishing.

Ethan

You're right. "Phase" is a real buzz-word in the world of audio.
A kind of inquisition tribunal for which all audible evils are adressed.
"Oh! I hear something bad...HaaaH! It must be a phase problem."
I ever heard a guy who was hearing a phase problem in an amplifier at 500Hz. He claimed few degrees.
This one is special. I also know a friend who bought a fourth power amplifier for a Linn tri-amped system.
He said this is a spare one in case of the company would modify future version of it.
So he would keep the same phase response in his system. Phase is a real buzzword and it works.

All this is known by the industry and is reflected in marketing.

Luke








----------
Sent via SPRACI - http://www.spraci.net/ - Parties,Raves,Clubs,Festivals
Bob Olhsson
2004-04-18 15:54:16 UTC
Permalink
Post by unknown
Ethan
You're right. "Phase" is a real buzz-word in the world of audio.
As is a consumer-grade "double blind testing" procedure that is so
insensitive that most people can't reliably tell the difference between a
digital recording and an analog cassette.

How about just getting rid of ALL the buzz-words and pseudo-science?
--
Bob Olhsson Audio Mastery, Nashville TN
Mastering, Audio for Picture, Mix Evaluation and Quality Control
Over 40 years making people sound better than they ever imagined!
615.385.8051 http://www.hyperback.com
Scott Dorsey
2004-04-18 16:25:49 UTC
Permalink
Post by Bob Olhsson
How about just getting rid of ALL the buzz-words and pseudo-science?
What fun would that be? The great thing about audio is just how obfuscated
everything is, and how it can be an adventure pulling a small grain of truth
out of a huge pile of marketing bullshit and misguided common knowledge.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Carey Carlan
2004-04-18 22:14:55 UTC
Permalink
Post by Scott Dorsey
Post by Bob Olhsson
How about just getting rid of ALL the buzz-words and pseudo-science?
What fun would that be? The great thing about audio is just how
obfuscated everything is, and how it can be an adventure pulling a
small grain of truth out of a huge pile of marketing bullshit and
misguided common knowledge. --scott
Eschew obfuscation.

--read that on a t-shirt somewhere.
unknown
2004-04-18 16:45:57 UTC
Permalink
How about just getting rid of ALL the buzz-words and pseudo-science?
Read Etha's article about Audio myths.
Maybe we could just suggest him some add-ons.

Luc





----------
Sent via SPRACI - http://www.spraci.net/ - Parties,Raves,Clubs,Festivals
unknown
2004-04-18 16:46:04 UTC
Permalink
How about just getting rid of ALL the buzz-words and pseudo-science?
Read Ethan's article about Audio myths.
Maybe we could just suggest him some add-ons.

Luc





----------
Sent via SPRACI - http://www.spraci.net/ - Parties,Raves,Clubs,Festivals
Arny Krueger
2004-04-18 19:02:22 UTC
Permalink
Post by Bob Olhsson
Post by unknown
Ethan
You're right. "Phase" is a real buzz-word in the world of audio.
As is a consumer-grade "double blind testing" procedure that is so
insensitive that most people can't reliably tell the difference
between a digital recording and an analog cassette.
Is this someone's invention, or did it really happen?
William Sommerwerck
2004-04-18 19:24:08 UTC
Permalink
Post by Arny Krueger
Post by Bob Olhsson
As is a consumer-grade "double blind testing" procedure
that is so insensitive that most people can't reliably tell the
difference between a digital recording and an analog cassette.
Is this someone's invention, or did it really happen?
The difference between a digital recording and a cassette recording -- at least
of a live performance -- should be audible within just a few seconds of
listening. I've made such a comparison, and the difference is hardly subtle.

On the other hand, most commercial recordings are so lousy that a good cassette
dub might be difficulty to quickly distinguish.
Nousaine
2004-04-18 19:53:40 UTC
Permalink
Post by William Sommerwerck
Post by Arny Krueger
Post by Bob Olhsson
As is a consumer-grade "double blind testing" procedure
that is so insensitive that most people can't reliably tell the
difference between a digital recording and an analog cassette.
Is this someone's invention, or did it really happen?
The difference between a digital recording and a cassette recording -- at least
of a live performance -- should be audible within just a few seconds of
listening. I've made such a comparison, and the difference is hardly subtle.
On the other hand, most commercial recordings are so lousy that a good cassette
dub might be difficulty to quickly distinguish.
In the early days of CD a tape company ran an exhibition test at an AES
Convention trying to show that people wouldn't be able to tell a CD from a
high-quality cassette. I can't recall the outcome.
Arny Krueger
2004-04-19 00:38:07 UTC
Permalink
Post by William Sommerwerck
Post by Arny Krueger
Post by Bob Olhsson
As is a consumer-grade "double blind testing" procedure
that is so insensitive that most people can't reliably tell the
difference between a digital recording and an analog cassette.
Is this someone's invention, or did it really happen?
The difference between a digital recording and a cassette recording
-- at least of a live performance -- should be audible within just a
few seconds of listening. I've made such a comparison, and the
difference is hardly subtle.
Of course I agree.
Post by William Sommerwerck
On the other hand, most commercial recordings are so lousy that a
good cassette dub might be difficulty to quickly distinguish.
Not in a close comparison.
William Sommerwerck
2004-04-19 01:47:24 UTC
Permalink
Post by Arny Krueger
Post by William Sommerwerck
On the other hand, most commercial recordings are
so lousy that a good cassette dub might be difficult
to quickly distinguish.
Not in a close comparison.
Note what I said! A close comparison usually requires "extended" listening.
Arny Krueger
2004-04-19 15:50:01 UTC
Permalink
Post by William Sommerwerck
Post by Arny Krueger
Post by William Sommerwerck
On the other hand, most commercial recordings are
so lousy that a good cassette dub might be difficult
to quickly distinguish.
Not in a close comparison.
Note what I said! A close comparison usually requires "extended" listening.
I disagree with that idea whole-heartedly. I don't see where a comparison
can be called "close" when the samples being listened to are greatly
displaced from each other in terms of time.

If I listen to something at 2 o'clock, and 3 o'clock, I don't think I am
going to be as sensitive to differences as I would be if I listened at 2
o'clock and 20 seconds later. Indeed, I know from years of experience that I
won't be as sensitive to differences with that kind of "extended" listening.

If I listen to something at 2 o'clock and 20 seconds later, I don't think I
am going to be as sensitive to differences as I would be if I listened at 2
o'clock and 2 seconds later. Indeed, I know from years of experience that I
won't be as sensitive to differences with that kind of "extended" listening.

I've looked carefully at listener sensitivity to differences that they can
hear, and find that putting the samples close together in the time domain
gives improved sensitivity. For example, if speakers are playing music at
two different levels in two different rooms, I might be able to hear a 1 dB
change. However, if I do a PCABX maximally-fast comparison, my threshold is
more like 0.4 dB.

The value I see to long term listening is that it seems to be a good way to
find candidates for critical passages that make differences as audible as
possible. IOW it's not the answer, but it can be a valuable way to figure
out how to ask the question.
William Sommerwerck
2004-04-19 17:50:34 UTC
Permalink
Post by Arny Krueger
Post by William Sommerwerck
Post by Arny Krueger
Post by William Sommerwerck
On the other hand, most commercial recordings are
so lousy that a good cassette dub might be difficult
to quickly distinguish.
Not in a close comparison.
Note what I said! A close comparison usually requires
"extended" listening.
I disagree with that idea whole-heartedly. I don't see where a comparison
can be called "close" when the samples being listened to are greatly
displaced from each other in terms of time.
Halt! Time out! We're getting into semantics.

Why do people have to spell out everything in detail?

When I said "quickly distinguish," I meant listening for five or six seconds --
the amount of time in which one can easily distinguish between digital and
cassette master recordings of live sound. I'm not talking about listening for
half an hour before switching, or listening at different times.

If I ever get my essay "Science -- Threat or Menace?" completed, I will address
the issue of long- and short-term listening.
Arny Krueger
2004-04-19 18:00:08 UTC
Permalink
Post by William Sommerwerck
Post by Arny Krueger
Post by William Sommerwerck
Post by Arny Krueger
Post by William Sommerwerck
On the other hand, most commercial recordings are
so lousy that a good cassette dub might be difficult
to quickly distinguish.
Not in a close comparison.
Note what I said! A close comparison usually requires
"extended" listening.
I disagree with that idea whole-heartedly. I don't see where a
comparison can be called "close" when the samples being listened to
are greatly displaced from each other in terms of time.
Halt! Time out! We're getting into semantics.
Ironically, you defined "quickly distinguish" which you used two posts back,
but you didn't define "extended" which you just used.
Post by William Sommerwerck
Why do people have to spell out everything in detail?
We've got some well- loaded words here. "extended" as applied to listening
tests is one of them.
Post by William Sommerwerck
When I said "quickly distinguish," I meant listening for five or six
seconds -- the amount of time in which one can easily distinguish
between digital and cassette master recordings of live sound.
Agreed.
Post by William Sommerwerck
I'm not
talking about listening for half an hour before switching, or
listening at different times.
Good.
Post by William Sommerwerck
If I ever get my essay "Science -- Threat or Menace?" completed, I
will address the issue of long- and short-term listening.
Let me suggest that the threat or menace is not Science, but rather the
abuse and/or ignorance of it.
William Sommerwerck
2004-04-19 18:08:48 UTC
Permalink
Post by Arny Krueger
Post by William Sommerwerck
If I ever get my essay "Science -- Threat or Menace?" completed,
I will address the issue of long- and short-term listening.
Let me suggest that the threat or menace is not Science, but rather
the abuse and/or ignorance of it.
My tongue was in my cheek.
Arny Krueger
2004-04-19 18:48:15 UTC
Permalink
Post by William Sommerwerck
Post by Arny Krueger
Post by William Sommerwerck
If I ever get my essay "Science -- Threat or Menace?" completed,
I will address the issue of long- and short-term listening.
Let me suggest that the threat or menace is not Science, but rather
the abuse and/or ignorance of it.
My tongue was in my cheek.
Pardon my misspent Usenet "youth". ;-)

Your comment would be and has been taken quite seriously at RAO and RAHE.
Scott Dorsey
2004-04-19 02:08:15 UTC
Permalink
Post by Arny Krueger
Post by William Sommerwerck
On the other hand, most commercial recordings are so lousy that a
good cassette dub might be difficulty to quickly distinguish.
Not in a close comparison.
I can build you a speaker system so bad that you'll have trouble telling
the difference between an original master tape of the Rolling Stones from
a Mantovani cassette.

No matter how obvious a difference seems, I can build a bad enough test
to obscure it.

This skill is not a unique one; many demo rooms at the AES show can be
found demonstrating such things.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Jerry Steiger
2004-04-21 03:36:16 UTC
Permalink
Post by Scott Dorsey
I can build you a speaker system so bad that you'll have trouble telling
the difference between an original master tape of the Rolling Stones from
a Mantovani cassette.
Thank you, Scott, we needed that!

Jerry Steiger

Mike Rivers
2004-04-18 19:54:19 UTC
Permalink
Post by Bob Olhsson
How about just getting rid of ALL the buzz-words and pseudo-science?
What? And put thousands of marketeers out of work?

--
I'm really Mike Rivers (***@d-and-d.com)
However, until the spam goes away or Hell freezes over,
lots of IP addresses are blocked from this system. If
you e-mail me and it bounces, use your secret decoder ring
and reach me here: double-m-eleven-double-zero at yahoo
Bob Cain
2004-04-18 21:49:41 UTC
Permalink
Post by unknown
Ethan
You're right. "Phase" is a real buzz-word in the world of audio.
And for the most part irrelevant. Phase was and is
meaningful in the context of feedback based systems but for
audio it's all about its derivative, group delay.

Linear phase is nowhere as meaningful to the ear as its
proponants would have us believe. We spend a lifetime
interpreting sound that is delayed and time dispersed in all
kinds of ways and have pretty damned good mechanisms built
in and adapted for subjectively ignoring it.

However, linear phase filters present to that adapted
mechanism a form of delay and time dispersion that nature
never ever produces so it has no experience in dealing with it.


Bob
--
"Things should be described as simply as possible, but no
simpler."

A. Einstein
William Sommerwerck
2004-04-19 01:50:34 UTC
Permalink
Post by Bob Cain
Linear phase is nowhere as meaningful to the ear as its
proponents would have us believe. We spend a lifetime
interpreting sound that is delayed and time dispersed in
all kinds of ways and have pretty damned good
mechanisms built in and adapted for subjectively ignoring it.
You're misusing the term "linear phase." A true linear-phase filter has constant
group delay, and should not audibly alter the sound.

As I said, the issue is whether non-constant group delay is audible -- not
"phase."
Post by Bob Cain
However, linear phase filters present to that adapted
mechanism a form of delay and time dispersion that nature
never ever produces so it has no experience in dealing with it.
Linear-phase filters do not introduce dispersion or relative delay.
Bob Cain
2004-04-19 04:11:40 UTC
Permalink
Post by William Sommerwerck
You're misusing the term "linear phase." A true linear-phase filter has constant
group delay, and should not audibly alter the sound.
Sorry, but that's wrong. I can show you a bass drum hit
processed with a linear phase bass boost filter where the
low frequency begins appearing signifigantly before the
percussive hit that starts it all. I find this audible.

All the consequences of linear phase are very poorly
understood. Yes, the group delay can be kept constant while
modifying the amplitude but the price is that a frequency
component can be seriously spread out or smeared about the
center that determines its average delay.

Linear phase is just not for listening to. It does
something to the sound that nature cannot do. Nature is
strictly causal. Linear phase filtering is not.



Bob
--
"Things should be described as simply as possible, but no
simpler."

A. Einstein
William Sommerwerck
2004-04-19 14:11:34 UTC
Permalink
Post by Bob Cain
Post by William Sommerwerck
You're misusing the term "linear phase." A true linear-phase
filter has constant group delay, and should not audibly alter
the sound.
Sorry, but that's wrong. I can show you a bass drum hit
processed with a linear phase bass boost filter where the
low frequency begins appearing signifigantly before the
percussive hit that starts it all. I find this audible.
Then it's not a linear-phase filter.

You're probably confusing linear phase with minimum phase. They're not at all
the same thing.
Post by Bob Cain
Linear phase is just not for listening to. t does
something to the sound that nature cannot do.
Nature is strictly causal. Linear phase filtering is not.
It sounds as if you're talking about an IIR filter, which can be "non-causal"
(ie, the output can "precede" the input). (Strictly speaking, this is
impossible. I'm thinking of "pre-ringing.")
Arny Krueger
2004-04-19 16:21:27 UTC
Permalink
Post by William Sommerwerck
Post by Bob Cain
Post by William Sommerwerck
You're misusing the term "linear phase." A true linear-phase
filter has constant group delay, and should not audibly alter
the sound.
Sorry, but that's wrong. I can show you a bass drum hit
processed with a linear phase bass boost filter where the
low frequency begins appearing significantly before the
percussive hit that starts it all. I find this audible.
Then it's not a linear-phase filter.
It seems hard to explain how a linear phase filter would affect the timing
of a signal in a way that differs from that of a pure delay. That means that
sound at any frequency would have to be reproduced with the same timing as
in the source.

However, I've got to pinch myself and remember that the phrase " linear
phase" describes the phase response of the filter. I don't think it says
anything about the amplitude response. I think your comments agree with this
idea, as far as they are stated.

However, a linear phase filter with non-flat frequency response can

(a) exist and
(b) have audible effects.

It just can't mess with timing of signals. It can mess with the relative
amplitudes. If you change the relative amplitude of the components of a
waveform, you can definitely distort the appearance of the waveform.
William Sommerwerck
2004-04-19 17:56:47 UTC
Permalink
Post by Arny Krueger
It seems hard to explain how a linear phase filter would affect
the timing of a signal in a way that differs from that of a pure
delay. That means that sound at any frequency would have to
be reproduced with the same timing as in the source.
Correct.

To be linear phase, a filter's phase shift must be directly proportional to
frequency. In addition the plot of phase vs. frequency must pass through the
origin (0,0). That is, the plot (at least over the frequencies of interest) must
be representable as phi = kf, not phi = kf + c.
Post by Arny Krueger
However, I've got to pinch myself and remember that the phrase
"linear phase" describes the phase response of the filter. I don't
think it says anything about the amplitude response. I think your
comments agree with this idea, as far as they are stated.
Also correct.
Post by Arny Krueger
However, a linear phase filter with non-flat frequency response can
(a) exist and
(b) have audible effects.
It just can't mess with timing of signals. It can mess with the relative
amplitudes. If you change the relative amplitude of the components of a
waveform, you can definitely distort the appearance of the waveform.
Thrice correct.

The kind of filter you're talking about would normally start as a design that
introduces a particular frequency-response alteration, then add non-linear group
delay to linearize the group delay of the filter as a whole.

An example of such a design would be adding group delay to the equalizer for an
analog audio tape recorder, to compensate for the fact that HF spacing loss has
no phase shift associated with it, but the equalization that corrects for it
does.
Bob Cain
2004-04-19 16:28:01 UTC
Permalink
Post by William Sommerwerck
Post by Bob Cain
Post by William Sommerwerck
You're misusing the term "linear phase." A true linear-phase
filter has constant group delay, and should not audibly alter
the sound.
Sorry, but that's wrong. I can show you a bass drum hit
processed with a linear phase bass boost filter where the
low frequency begins appearing signifigantly before the
percussive hit that starts it all. I find this audible.
Then it's not a linear-phase filter.
Since I created it, I'm pretty damn sure. FWIW, audio DSP
is my schtick and I'm well educated in theory and practice.
Post by William Sommerwerck
You're probably confusing linear phase with minimum phase. They're not at all
the same thing.
I fully understand the difference and how to go between
them. If you wish I can send you the Matlab programs I've
written that compute minimum phase or linear phase versions
of a filter from any FIR specification. That is, the
results will have the same magnitude as the specification
but with the desired phase relationship.
Post by William Sommerwerck
Post by Bob Cain
Linear phase is just not for listening to. t does
something to the sound that nature cannot do.
Nature is strictly causal. Linear phase filtering is not.
It sounds as if you're talking about an IIR filter, which can be "non-causal"
(ie, the output can "precede" the input). (Strictly speaking, this is
impossible. I'm thinking of "pre-ringing.")
I think you've got considerations of the two backwards. An
IIR filter is purely causal as is any minimum phase filter.
It is the FIR linear phase that has the "pre-ringing" and
if you remove the fixed delay involved in it (half the
length of the filter) then it is symmetrically acausal.


Bob
--
"Things should be described as simply as possible, but no
simpler."

A. Einstein
William Sommerwerck
2004-04-19 18:03:36 UTC
Permalink
Since I created it, I'm pretty damn sure. FWIW, audio DSP
is my schtick and I'm well educated in theory and practice.
Uh, oh...
I think you've got considerations of the two backwards.
An IIR filter is purely causal as is any minimum phase filter.
It is the FIR linear phase that has the "pre-ringing" and
if you remove the fixed delay involved in it (half the
length of the filter) then it is symmetrically acausal.
An FIR filter is simply the sampled-data implementation of the filter's impulse
response with respect to its Laplace transform. (I know, because I invented the
FIR about 30 years ago -- then discovered Blumlein had beaten me to it 30 years
earlier.)

Because an IIR filter uses feedback, it can be non-causal.

I think I'm right, but I'd better stop at this point and pull out some reference
books. If I can find them among the junk...
dan lavry
2004-04-19 23:18:41 UTC
Permalink
Post by William Sommerwerck
Post by Bob Cain
something to the sound that nature cannot do.
Nature is strictly causal. Linear phase filtering is not.
It sounds as if you're talking about an IIR filter, which can be "non-causal"
(ie, the output can "precede" the input). (Strictly speaking, this is
impossible. I'm thinking of "pre-ringing.")
I think there is a lot of confusion here.

Linear phase means that all frequency components are delayed by the
same amount of time. So a linear phase system, assuming no EQ (all
frequencies passed equaly) yields the SAME WAVEFORM output as what
came in.

Analog filters (such as EQ) can not provide you with attenuation or
boost without ALSO AT THE SAME TIME altering the phase response. In
fact the "rules of the game" are somewhat stiff - analog circuits work
with poles and zeros. You insert a pole at a frequency f0, and it is
where you will have 3dB less gain. The slop will be 6dB per octave,
and the phase shift 45 degrees. The phase alteration due to that pole
at an ovtave below (f0/2) is near zero. At an octave above it is near
90 degrees.
The bottom line is - analog filtering alters phase. It is not phase
linear.

The IRR is a digital aproximation of analog filters done in the
digital domain. They are very efficient compute power wise. One must
be doing the design right to avoide art-effects (such as limit
cycles). There is a lot to say here and I will just leave it alone for
now.

One can do a linear phase EQ in the digital domain - the FIR filter.
FIR's is where the pre ringing stuff comes in, and you may hear it
with some particular audio signals if the sample rate is too low. By
the time the sample rate is 60KHz you are certainly away from such
problems.
FIR works great for audio with 88.2KHz and 96KHz sampling. It takes a
lot of processing power to do well.

FIR is non causal. I would not call IIR's not causul. Normaly, you do
not get an output out of IIR or an analog circuit before the impulse.
You do get preringing out of FIR's. I think here is too much talk
about preringing this days, it should be quantified. Some low
amplitude ringing at a frequency above your hearing is not the key to
everything, in my view. You may hear it with 44.1KHz but not at
88.2KHz... And it should be weighed against the advatages of FIR's
which are many.

Back to hearing phase - it has been said that if all audio channels
have the same phase curve, the sound arrives at your ears at the same
time for any given frequency, so "all is fine". The claim here is that
linear phase is an overkill and "equal phase response" treatment for
all channels is the key. I do not subscribe to it because changing the
waveform (non linear phase) will be treated differently by the
electronics. For example, I can take a fast rising signal and alter
the phase to lower slope... In any case, I do not know many that do
the same EQ to all channels, especialy in the case of many tracks...

BR
Dan Lavry
Lavry Engineering
Nousaine
2004-04-18 19:49:03 UTC
Permalink
Post by unknown
Luke,
Post by unknown
What David Clarke humbly tried to prove is only that phase audibility is
not a first order priority and non-linear magnitude frequency response is
much more relevant. <
That's a beautiful one-sentence summary, and I agree completely.
It really doesn't matter to me whether extreme amounts of phase shift are
audible. It's not much of a problem in the overall scheme of things. The
same could be said for ultra-high sample rates and bit depths, the value of
passive summing, and a dozen other such things that are discussed far more
than deserved.
Too many people who have *no idea* how audio really works blame phase shift
because they continually read magazine and web articles saying phase shift
is a problem. Interviews with their favorite artists and producers parrot
the same words. I've been following this nonsense for 25+ years now. I first
noticed "phase as mystique" in the ads for the original Aphex Aural Exciter,
which was total BS and lied saying it's based on phase shift. It's a
glorified fuzz box, fer cryin' out loud, but I guess that doesn't have the
right marketing pizzazz.
It's been all downhill since then.
--Ethan
David Clark would alsopoint out that any phase anomaly large enough to cause
audible problems will also show a related anomaly in frequency response.
Arny Krueger
2004-04-17 15:33:27 UTC
Permalink
Post by Ethan Winer
Scott,
there are a lot of speakers on which it is clear that something is just
wrong in the crossover region. That something might just be
destructive interference issues off-axis, but it might also be total
group delay on-axis. <
This is the main point Dave makes in the first part of his article.
Phase shift from the crossover, or from the drivers' physical
alignment, can clearly affect the sound. But the real culprit is a
skewed *frequency response* caused when the drivers' outputs combine
in the air. Not the phase shift itself.
Agreed.

Deferential phase shift exists whenever two or more sources are combined.

This can happen in electronic circuits. For example, a console has send
buses and returns. Two or more sources are combined when the return is
mixed back in. When the devices that are connected between the send bus and
returns have phase shift or time delay, we get highly skewed frequency
response. It adds to any frequency response changes caused by those devices.
It is caused when two or more sources that have different phasing combine in
the mixer.

In acoustical situations this all happens all the time. If nothing else, the
second, third, fourth and so on sources, are simply reflection(s) of the
original source. There is zero chance that there is no phase shift and/or
time delays.

The effects of differential phase shift on frequency response are generally
strong, and are often catastrophic. The results are, first and foremost,
relatively large frequency response variations. The sensitivity of the ear
to relatively small changes in frequency response is remarkably high. The
ear's sensitivity to large changes in frequency response is of course, even
greater. While there is a controversy about the audibility of small amounts
of phase shift, there is no controversy about the audidibility of large
frequency response variations.
Loading...