Beyond the Ariel

[gedlee]

Originally posted by DDF

There's a wealth of rigorous testing info showing the audibility of moderate group delay with headphones, but just not in room.

A reference please.

[DDF]

Originally posted by gedlee



A reference please.

Something I wrote 15 years ago in the post to follow. I can update it, but am short on time. Of special note, I direct your attention to the audition studies on the interplay of group delay/phase and harmonics. Some examples:

From "Hearing, Its Psychology and Physiology" by the
Acoustical Society of America (my fave reference) on the
audibility of 2nd harmonic distortion with a pure 370
Hz fundamental:

"The masking of an added harmonic is negligible below
a sensation level of 40 - 50 dB. From 50 to 80 dB,
the amount of harmonic necessary for an audible change
increases rapidly, first in absolute magnitude, and later
in relative magnitude as well...."

"...The qualitative character of the audible change produced
by adding this harmonic was different at the various sensation
levels of the fundamental. At low levels the harmonic was usually
heard as a separate tone. In the middle region [50 - 80 dBSPL]
it was heard as a sharpening or brightening of the timbre of the
tone, whereas at high levels the changes were so complex and so
dependent upon differences of phase that any generalization about
their character would be misleading."

From "The representation of speech in the peripheral auditory
system", there's a paper by Schroeder and Mehgart "Auditory Masking
in the perception of speech", where they show that reversing the
phase of even one harmonics component is audible. He could even
"produce little melodies by sheer phase manipulations."

Summarizing:
Some monaural phase effects can be explained by the concept of the
inner spectrum, the spectrum available to the inner ear. This is
different than the spectrum at the outer ear due to non linearities
in the middle ear and inner ear. Identical external power spectra
can lead to substantially different inner spectra for different
phase angles


From "Sensation and perception", pg. 85:

"sounds whose time arrival differs by as little as 0.1 ms (no
intensity differences) are sufficient to serve as cues for
localizing sound in space (Rosenzweig, 1961).

This implies that for non coincident drivers smearing of a complex
signal in space can occur if the group delay at one frequency
exceeds that of another frequency by more than 0.1 ms.

[DDF]

Originally posted by DDF



Something I wrote 15 years ago in the post to follow.

>Dave,
>Any idea what this range of phase delay is for the audio band? I guess you
>mean that phase delay is more audible at some frequencies than others. Can
>you give endpoints and/or ranges?
>
>Frequency Most can hear it Some can maybe hear it
> Hz

Kirk, I wish things were so simple. Researchers have been wrestling with
the question of group delay audibility hot and heavy for the past 15 years
and a large body of research is amassing. I'll quote a great deal of it
here. So, grab a coolie, this'll take few minutes...8

First lets agree on our definitions. Your question refers to the
acceptable range of "phase" delay. Well, let's change your question
to "What is the acceptable range for "Group" delay?"

Phase delay is actually comprised of two components.

First, there is a linear phase slope that can be removed from
our system's phase response. Since "delay" is mathematically
the negative slope of phase vs. radian frequency, removing a
linear phase component is the same as removing a constant
time delay applying across the entire frequency band.

This is OK. Since audio reproduction is a simplex process,
we don't care what our constant delay is.

This brings us to the second component of phase delay,
which is Group delay. This the relative delay between
frequencies, with the delay at one frequency taken as the
reference value. The reference is usually taken as the
lowest delay, that way our group delay always remains positive.
This is conceptually easier for our pea brains to grapple with.

OK. Heavy breath. Plunge:

First off, every research paper I've ever read has stated that
slight changes in amplitude response are far more audible than
slight changes in phase response. The lesson is that one
should not sacrifice flatness in amplitude response solely for
gains in phase response linearity.

That said, lets go on.

If our system were truly minimum phase, a flat amplitude
response would automatically result in a linear phase
response. Loudspeaker systems are almost never minimum
phase due mainly to the common usage of non-coincident radiators
i.e. woofers and tweeters with apparent acoustic sources
seperated in space.

Many studies have been performed where the reasearchers
have taken a non-minimum phase loudspeaker system and
applied minimum phase all-pass equalizations to determine the
audibility of different group delays introduced by the all-pass
filters.

Lipshitz showed (JAES, around '85) that a non-minimum phase
system can not possibly be made minimum phase through
such application of minimum phase equalization. Non-minimum
phase eq must be used. This shows that audibilty experiments
of group delay are very hard to perform with loudspeakers since
a constant group delay is very hard to arrive at without using
DSP techniques (where phase and amplitude eq can be handled
independantly).

Research also shows that it is far easier to hear group delay
differences on headphones than loudspeakers, one reason being that
they tend to be much more near minimum phase and the second being
that room reflections which could mask perceptions of delay
are avoided.

The results I'm about to quote were taken with headphones.

From "Phase Distortin and Phase EQ in Audio Signal
Processing- a Tutorial Review" By Doug Preis, Tufts Univ. and
delivered at the 70th convention of the AES, Oct. 1981:

"Perceptual thresholds for detection of group delay distortion
depend on the test signal used, the method of irradiation,
and the training of the auditor."

He then proceeds to graphically represent the results of 7
studies, each using headphones and each using non-musical signals
(clicks, sines, tone bursts, etc.). Each test was performed
over a narrow frequency band. He then formed a composite
template based on these results. This is shown graphically below
with the vertical axis being the audibility threshold
for group delay (milliseconds).


3 |
| ********
2 | *
| *
1 | ******
|
--------------------------------
|
-1 | ******
| *
-2 | *
| ********
-3 |
|
-------------------------------
.05 .1 .2 .4 1 2 4 10 20

Frequency (kHz)

Again quoting:

"the ...results indicate that large variations (few ms) are permissible
in the low frequency range wheras variations exceeding 0.5 ms
in the 1 to 5 kHz range may be perceptible under
sensitive test conditions." He then makes an unsubstantiated guess
that the detection thresholds of group delays for loudspeakers in rooms
will be at least twice that shown by the chart. I think he was being a
bit charitable and that it's actually much higher.

"Variations within these bounds should be imperceptible".

"(These) tolerances are not directly applicable to speech or music
signals irradiated by loudpeakers in a reverberant environment"

He also points out that more work is needed to extend the data below
and above the range shown.

Moving on to "Perception of Phase Distortion in
All-Pass Filters", by J.A. Deer, P.J. Bloom and D. Preis as
printed in the JAES, Oct, 1985:

"Results from listening tests indicate that a statistically significant
perceptual threshold is reached when peak group delay distortion at
2 kHz is in the neighborhood of 2 ms (for diotic presentation via
earphones)."

From S.P. Lipshitz, M Pocock and J. Vanderkooy, "On the audibility
of Midrange Phase Distortion in Audio Systems", JAES, Sept. 1982:

"An argument frequently put forward to justify why phase distortion
cannot be significant for material recorded and/or reproduced in
reverberant surroundings is that reflections cause gross phase
distortions themselves, which are very position sensitive. This is
true, but in both cases the first arrival direct sound is not subject
to these distortions, and very important directional and other
analyses are conducted during the first few milliseconds after its
arrival, and before the predominant reverberation arrives."

But, they then go on to discuss that they had a much harder time
detecting delays in such an environment relative to headphone use.

They also discuss papers from the past on delay audibility and also
papers detailing how the ear works and why it can hear phase changes,
then throw in a little politics at the end of their paper:

"We do not understand why there are still reports appearing which
state that the human ear is deaf to non-linear phase change"

Laurie Fincham of KEF fame tried to tackle the issue of low
frequency audibility in "The Subjective Importance of Uniform
Group Delay at Low Frequencies", JAES, June '85:

He studied the effect of the typical analog record chain on low
frequency phase performance and its perceptual results:

"Listening tests in carefully controlled conditions indicate that
a reduction in group delay distortion at low frequencies in
the replay chain is probably worthwile only when the recorded
material is itself also free from such distortions. These
effects are however quite subtle."

He did find that the insertion of two cascaded all-pass filters
with a Q of root(2) each (Butterworth), and maximum phase shift
at 40-50 Hz did in fact "cause distinct audible differences to be
observed by most of the audience in a typical lecture theatre".

A question I ponder is: did they hear the group delay or
distortions from the circuit used to implement it?

To close this epic, I'll quote from F.E. Toole, from
"Loudspeaker Measurements and their relationship to Listener
Preferences: Part 1", JAES, April "86:

"Using music, and listening to loudspeakers in normal rooms,
even carefully selected listeners appear to have difficulty
detecting the presence of quite large phase shifts, much
less are they able to establish a preference".

"The recent work has been more thoroughly investigative and puts
limits on the thresholds of audibility of various phase and
group delay effects. The limits on over-all trends in
phase response are very generous, and appear not to require special
consideration in the design of conventional domestic
loudpeakers." He then points out that phase shifts associated
with resonances should be minimized by controlling the resonance.

What does all this tell us:

1. We don't know a great deal more that we know;
2. Group delays of 0.5 ms are audible from 1 to 5 kHz only
under extremely artificial conditions: heaphone use, simple
sources (clicks, sines) and acute listeners.
3. It is much harder to detect group delay differences using complex
sources (music and speech). These thresholds have not been formally
determined (at least to my knowledge).
4. Low frequency group delay distorions in systems seem to be swamped
by low frequency group delay distortions in most analog record
systems. Digital record systems fair much better, so reductions
in low frequency group delay distortion in loudspeakers may be
worthwile, as most new recordings are done digitally.
5. My available research can not concisely answer Kirk's question
in the context he asked it, but we do know the effects are audible
to an extent with headphones, and to a lesser and unknown extent
with loudspeakers.

Better get back to work before I'm sacked....

Dave Dal Farra
Still at BNR Ottawa
Audio and Acoustics Group

[john k...]

Earl,

With regard to the impulse smoothing and the ICTA crossover, they are two separate topics that have entered the mix.

With regard to the GD arguments I would say most doesn't equally all, not to mention that a negative result can't be proven. I do accept that there are many instants where GD variation across the audio band is not heard, at least as such. But it is still a form of linear distortion in the time domain and in that regard my desire from an engineering and audiophile point of view is that if it can be eliminated without introducing other adverse effect then it should be eliminated. It's a design objective and if I can achieve good power response and dispersion, at least as good as any other high quality speaker, then why not?

Regarding impulse smoothing, as you see I did this back in May when I was aiding the developer of SoundEasy with cepstral editing. My feeling was that direct editing of the impulse was just as attractive or better, and the write up was to demonstrate why I felt that way. In any event, it's not of particularly great interest to me in that I believe merging the far field response with the low frequency near field data, corrected for the baffle 2Pi to 4PI transition is superior since only the 2Pi to 4Pi baffle step effect need be modeled. And I believe the baffle step effect can be adequately modeled. Baffle shape and edge treatment play little roll once the wave length is much larger than the baffle characteristic dimension.

[gedlee]

All interesting comments but they still point me in the direction that most research says that GD is not a major issue. And no I don't agree that correcting something for correction sake is a worthy exorcize.

My research involved mostly non-minimum phase GD and showed it to be a far more significant effect than minimum phase GD has shown. In fact my research contradicted Moores if you don't consider that Moore did his work at a fairly low signal level. The fact is that GD, or at least non-minimum phase GD is an audible effect if one considers playback level. I think that it is this level component that was missing in most previous work and the reason why there are contradictor results.

I think that in a GD discussion we must consider two different aspects: 1) the playback level 2) if the GD is minimum phase or not. Seldom if ever are these two factors controlled or studied as factors in and of themsleves.

DDF - thanks for posting those studies. I see relavence in some but not all. There has been some substantial recent work - most notably the Moore papers, which you should also be quoting.

[DDF]

Originally posted by gedlee


I think that in a GD discussion we must consider two different aspects: 1) the playback level 2) if the GD is minimum phase or not. Seldom if ever are these two factors controlled or studied as factors in and of themsleves.

One root cause here appears to be the interplay of the signal harmonic content and the ear's non-linearity. The ear's non-linearity is time variant.

I'd also add a signal content aspect. Some research shows heightened GD audibility when the signal is harmonically complex, and the envelope has some "beating" to it. Theory being that the GD allows previously masked components to unmask at periods of time when the amplitude envelope shrinks.

I find it most interesting that your research showed distortion to have less relevance than previously assumed: traditional GD reduction often comes at the cost of increased driver distortion (lower slopes).

[gedlee]

Originally posted by DDF


I'd also add a signal content aspect. Some research shows heightened GD audibility when the signal is harmonically complex, and the envelope has some "beating" to it. Theory being that the GD allows previously masked components to unmask at periods of time when the amplitude envelope shrinks.

I find it most interesting that your research showed distortion to have less relevance than previously assumed: traditional GD reduction often comes at the cost of increased driver distortion (lower slopes).

Personally I think that the use of any signal but real music leads to erroneous or at least pointless results. I'm a practical person who seeks to improve the listening experience of music not sine waves. Specialized test signals are useful for people digging deep in the science of hearing, but as a loudspeaker designer its not too relevent. Like I said to John, I am not big on fixing things that don't matter. There is so much wrong that does matter that chasing the unimportant just does not seem practical to me.

Thats why I am now of the belief that nonlinear distortion in loudspeakers is irrelavent. Not that I can't make a loudspeaker where nonlinearity is factor, the point is that I can make one where it isn't. I prefer to do the latter rather than the former.

I've been designing audio products for so long that I am simply not interested in doing, studying or arguing about something that isn't a significant audible factor - and this latter point has to be proven to me in the first place.

For years I studied nonlinear distortion in loudspeakers - the whys, and wherefors (you can see this in many of my publications from about a decade ago). Then when I actually sat down to find out how important it was I found that it wasn't - what a waste of time. "I wish that I had know then what I know now!"

[Lynn Olson]

Thanks for the discussion, Dr. Geddes, JohnK, and DDF. All of it at a higher level of rigor than I usually use, guessing and fumbling along, following my nose most of the time, with the occasional insight or two.

The discussion also shows why serious designers, who've been doing this a long time, agree on some matters, but not others. Read carefully enough, set aside the disagreements for a moment, and discover there are nuggets of illumination everywhere.

I'm a fan of James Boyk, a former professor at CalTech, and here are some of my favorite articles:

On Both Sides of the Microphone

Capturing Music: The Impossible Task

Small-Signal Distortion in Feedback Amplifiers (PDF)

Audiences of the World, Arise!

The Ear of the Beholder, with the following memorable quote:

"This reminded me of descriptions of concerts I sometimes hear from friends. They will tell me in a dutiful tone what famous performer they heard, how wonderful his technique was, and so on. This always makes me suspicious, and I generally find that when I ask if the concert moved them, they look a little surprised and uncomfortable at my Midwestern naivete - you mean that's your criterion? - and then they answer, No.
In a way, though, you should not notice the performance at a concert. The music should be transparent to the emotion. And a reproduction system should be transparent to the music.

People accept amazing ugliness in reproduced sound because it's impressive or overwhelming, or because it emanates from expensive or reputedly good equipment. In the long run it all comes out in the wash; but as Keynes said, in the long run we are all dead.

Meanwhile, you may be the live owner of equipment you can't stand. To short-circuit this process, listen at some length before you buy. Notice your state of muscle tension, notice how quickly you tire of the sound, notice whether you are attending to the equipment or the music. Ask yourself if the sound is beautiful. That is more important than volumes of technical data!"

[EdwardWest]

Originally posted by Lynn Olson
...I'm a fan of James Boyk, a former professor at CalTech...

Lynn,

Thanks for the links. I particularly like this one from Dynamic Inflection and the Beauty of Live Music:

But unlike the situation with solid-state amplifiers, where I have at least heard a couple of absolutely top-quality units, with digital I have heard none at all, even after several years of trying. These perceptions are the reason that we do All-Tube AnalogTM recording. They also underlie our hyperbole, "Digital finishes what the transistor began." - James Boyk, 1985

Also Lynn, could you send me an email. I sent a message to your 'nutshell' email address, but I am guessing you haven't received it.

Edward

[Lynn Olson]

E-mail received and acknowledged, you should have a reply by now.