Harman / Kardon: Loudspeakers and Rooms For Multichannel Audio Reproduction

The Science of Audio - a series of lectures by Floyd E. Toole, Ph.D.
Vice President Acoustical Engineering

Harman International Industries, Inc.
8500 Balboa Boulevard, Northridge, CA 91329 818 895 5761 [email protected]
harman / kardon
Loudspeakers and
Rooms for
Multichannel Audio
Reproduction
by
Floyd E. Toole
Vice President Acoustical Engineering
Harman International
Part 2 – Making a good loudspeaker -

Imaging, space and great sound in rooms.
Loudspeakers can be designed to be “room

friendly” so that they can sound good in a
variety of different rooms. Controlling
reflections can optimize imaging and spatial
effects.
Some of what we think we

Audio is an industry that is unfortunately burdened with a lot of ideas
know about audio is BS! and beliefs that are not based on physical facts. Our business is
Before Science complicated enough without adding the confusion of half-truths and
folklore. An understanding of at least the fundamentals of room
acoustics, and how loudspeakers and rooms interact with each other,
BS takes us a long way towards our goal of truly excellent sound. This
benefits everyone, the customer, the installer/consultant/retailer, and
the manufacturers of the equipment.
2
The fact is that audio is a technology with a firm scientific grounding

under most of it. It is not an art, and audio products do not possess
Music and Movies
artistic characteristics, except in the visual sense. The better we
are Art
understand and use the real science to achieve excellent performance
from our products, the more often the true audio artistry, the sound of
Audio is a Science music and movies, will be heard by us and by our customers.
The existence of custom audio installers and consultants is something

Science in the Service of Art relatively new to our industry. For the first time in the history of
is our Business!! audio, there are professionals whose job it is to help the customer, in
3
his or her own home, to achieve the best possible sound.
31 January, 2002 1
The Science of Audio - a series of lectures by Floyd E. Toole, Ph.D. Vice President Acoustical Engineering
The Goal: To deliver high The traditional problem in audio has been that the room, the final
quality sound to our customers’ audio component, is not within our control. Customer satisfaction,
ears. assuming that it is based on good sound, has been, therefore, a matter
The Problem: of chance.
ROOMS, the final audio component.
They affect sound quality and imaging

They dominate bass quality
This can change. With the selection of the appropriate loudspeakers,
They do this during the making of the recordings, and the application of some fundamental room acoustical knowledge and,
during their playback at home.
They are all different. if necessary, the right kind of equalization, we can greatly increase
the odds in the favor of the customer and thereby ourselves.
4
Loudspeakers should sound good . . . No matter what measurements tell us, a loudspeaker isn’t good until it
sounds good. Complications in determining what is “good” include
. . . and that is part of the problem. variations in rooms and recordings. The latter is something often
ignored as we go about our daily businesses.
How do we judge what is “good”?
When we listen we are instantly trapped in the “audio circle of

confusion”.
Loudspeakers are evaluated by listening to recordings.
We listen, of course . Recordings are made using microphones that are selected and
positioned, equalized and processed in a variety of ways using the
But when we do . . . masses of equipment in a recording studio.
All of this is done while listening through loudspeakers in a room – a
recording control room or movie dubbing stage.
The quality of the sound in a recording is very much dependent on the
quality of sound from the monitor loudspeakers in that particular
6
room.
The recording industry has NO critical standards relating to
The Audio Industry is in a “Circle of Confusion”
loudspeakers for monitoring and for the rooms in which they are
LOUDSPEAKERS
which are evaluated used. Consequently, recordings are extremely variable in quality,
by using
even in the gross characteristics of bass and treble balance. Yet, we
try to evaluate audio products using such recordings. It is like
MICROPHONES, EQ, RECORDINGS
REVERB & EFFECTS which are made making a technical measurement with an undefined test signal. The
which are evaluated by using
by using result is that mistakes are made. We cannot tell whether a good
sound is the result of a truly good loudspeaker/room combination, or
which are then used
to evaluate audio whether it is a case of compensating errors: a recording with, for
products
example, too much bass being combined with a playback system that
7
is deficient in bass.
Some control rooms sound superb, while others are back in the dark
ages of loud “mid-fi”. Some even go out of their way to use bad
Professional Monitor
LOUDSPEAKERS
which are evaluated
monitor loudspeakers that they think represent what people are
by using
listening to in their homes and cars. It is obvious to anyone who
listens carefully that all loudspeakers are getting better, and that the
MICROPHONES, EQ, RECORDINGS
REVERB & EFFECTS which are made good ones are sounding more and more alike, and much more like the
by using “real thing”. However, bad loudspeakers can be bad in an infinite
number of ways. No two are alike, and they can be dramatically
which are then
auditioned through
Consumer
different. How, then, is it possible for one “bad” monitor loudspeaker
LOUDSPEAKERS
to represent the huge variety of sounds from clock radios, boom
8
boxes, mini-systems, headphones and entry level car audio? It isn’t!!
31 January, 2002 2

All of us need to exercise whatever influence we have to elevate the
LOUDSPEAKERS
which are evaluated
quality of sound everywhere. Then, and only then, will we have
by using
some assurance, when we listen at home or in our cars, that we are
THE CREATION hearing what was intended by the artists. The enemy in this effort is
OF THE ART
MICROPHONES, EQ,
REVERB & EFFECTS
RECORDINGS
ignorance and apathy. Most customers are intimidated by these kinds
which are made
by using of decisions, and some truly say they don’t care. Yet, I have never in
my life demonstrated a truly good sound system to anyone who was
which are then
APPRECIATION
auditioned through not impressed, if not absolutely “blown away”.
OFConsumer
THE ART
LOUDSPEAKERS
9
LOUDSPEAKERS
Ironically, the problem exists at both the professional and consumer

levels. Both need to be aware of the genuine advances in acoustic
The “circle of confusion” science and technology.
is broken,
if these loudspeakers
sound like each other.
Consumer
LOUDSPEAKERS
10
LOUDSPEAKERS
PRESERVATION
The “circle of confusion” is broken,
Only then can we say that we are working within an industry that
aims to preserve the audio artistry.
OF THE ART
if these loudspeakers
have the same sonic signature -
i.e. if they are similarly good.
Consumer
LOUDSPEAKERS
11
This is a problem to which there If we cannot totally rely on our ears, what else is there?
is not a single, or a simple,
solution.
12
Science Involves Measurements, The scientific method requires data. Data of all kinds, and the more
and We Need to Do Two Kinds: the better. In this case, we must use technical measurements, because
OBJECTIVE SUBJECTIVE they are the essential tools of the engineers designing the products. It
1. Frequency Response - 1. Sound Quality - is necessary to measure everything that we think might be relevant to
on and off axis, spatial timbre, bandwidth
averages including 2. Directional and spatial how something sounds. This is more than is commonly thought.
sound power. effects in stereo and However, we also need subjective data, relating to listeners’ opinions
2. Phase Response multichannel systems
3. Non-Linear Distortion 3. Distortions and noises
of the many perceptual dimensions of sound quality as well as spatial
- THD, IMD, noise 4. Dynamic Capabilities and directional attributes.
4. Power Compression
13
31 January, 2002 3
If listeners like or dislike something, it is important to try to identify

Scientific Design Requires: what, technically, was responsible. By taking the listening reports
(a) carefully controlled listening tests: back to the lab, it is possible to learn the relationships between what
subjective measurements we measure, and what we hear. This is the science of
combined with psychoacoustics, and the better we understand it, the better we will be
(b) accurate and comprehensive at delivering good sound to our customers’ ears.
technical measurements
combined with Technical measurements must be accurate, or else they lie to us.
(c) knowledge of the psychoacoustic Getting accurate acoustical measurements without good facilities is
relationships between perceptions and very difficult to impossible. Much of the data floating around the
measurements .
loudspeaker industry is not accurate.
14
Most listening tests are valid only at a specific time and place, for the
What are Subjective Measurements? specific recordings that were listened to, and for the people offering
They are LISTENING TESTS in which the their opinions. This can be acceptable if it is you choosing your own
OPINIONS OF LISTENERS relate to the system in your own home. It is not acceptable for a loudspeaker
SOUNDS OF THE LOUDSPEAKERS rather than:
manufacturer, who is trying to design products that can sound good to
How they look
How much they cost These are many listeners, in many rooms, with many recordings.
What reviewers think of them NUISANCE Consequently, we get scientific about it, and start to remove some of
What listeners think of the manufacturer VARIABLES
How loud they are
the variables that have nothing to do with the sound from the
-they MUST be
How the room is physically arranged controlled loudspeaker, letting the listeners focus as much of their attention as
Etc. . . . possible on the sound, and the sound alone. Purveyors of ‘magic’ in
15
the audio industry find these double-blind tests very threatening.
Mono, Stereo and Multichannel One of the most problematic nuisance variables is the position of the
Tests loudspeaker and the position of the listener in the room. To control
this variable at Harman, we have created a room with a “shuffler” that
physically moves speakers around, bringing them always to the same
SHUFFLER
locations when they are being auditioned. It is pneumatically driven,

quiet, and computer controlled so that the positions are precise. It
takes about three seconds to switch the positions of the active and
inactive loudspeakers.
16
A pair of speakers ready for listening.
17
And, about three seconds later, a second pair can be heard.
18
31 January, 2002 4
The listener (and we prefer to use one listener at a time) sees none of
What the Listener Sees
this, of course. Here we show a video display on a large perforated
screen. For the evaluation of most products this is not used.
The tests are controlled by the listener, who takes as long as is needed
in order to form a satisfactory judgment. A computer randomizes the
choice of music, and the coded identity of the test loudspeakers for
each musical selection, so that the opinions must relate as much as
possible to the sound itself.
Listeners are selected for normal hearing and aptitude, and then are
trained to be really fussy. They yield remarkably consistent opinions.
19
“Where the rubber hits the road”, in the customer’s home, we have no
In the Real World
such conveniences, so we must develop products and techniques that
allow good sound to prevail even when the local acoustical conditions
The room is the final audio
are less than ideal.
component,
and it is not under control! This is where knowledgeable custom installers, consultants and audio
specialists come to the rescue.
20
The Room - Direct Sound
The first sound to arrive at a listener’s ears is the “direct” sound. If

the loudspeakers have been angled to face the listener, this will be the
on-axis sound, often the best possible sound from the loudspeaker.
21
The Room - Early Reflections

However, following only a few milliseconds behind, and only slightly
less loud, will be the early reflections: sounds that have been reflected
from only one surface in the room.
22
Still later, come the multitudes of reflections that have been reflected
The Room - Reverberation
more than once, perhaps many times. These are individually much
lower in amplitude, but collectively loud enough to be a powerful
factor in our impressions of sound quality, space and imaging. In
small rooms, typically furnished, this sound field, although often
called ‘reverberation’ is not the directionally diffuse and temporally
complex reverberation that we hear in a concert hall, or many other
large, acoustically ‘live’ spaces. Some would argue that it deserves a
different name.
23
31 January, 2002 5
The Room – Resonant Modes

Rooms also have resonances that emphasize certain frequencies,
attenuate others, depending on the dimensions and shape of the room.
And, they do so in a manner that depends on where the loudspeakers
LENGTH and listeners are located within the room. These effects are strongest
WIDTH at low frequencies.
HEIGHT
24
A Problem Loudspeaker Using a loudspeaker that we know, in retrospect, had a design flaw,
10 let us see what happens in a room. This loudspeaker was designed
woofer midrange tweeter
using the philosophy that the direct sound, the on-axis sound, is the
0
dB most important. The top curve is the on-axis measurement, and it is
-10 ON AXIS very smooth and flat, a credit to the design team.
30 DEGREES OFF AXIS The second and third curves, moving downward, are the 30- and 60-
-20 60 DEGREES OFF AXIS
degree off-axis measurements, showing that these sounds are not
nearly so neutral; the output varies with frequency. What happens to
20 50 100 500 1K 5K 10K 20K
FREQUENCY (Hz) this in a room?
25
The data in this slide are derived from many measurements made in a
What Happens in a Room? large anechoic chamber. This is a room having no “echoes”, used for
20 acoustical measurements. All surfaces are covered with highly
PREDICTED“ROOM CURVE”
effective acoustical absorbing material that is, in this case, about four-
10
dB
feet thick. The color coding of the curves is not visible here, so it is
0 not possible to see which curve is which. However, one might
DIRECT SOUND recognize the flat on-axis curve representing the direct sound, and see
-10
EARLY REFLECTIONS also that none of the other curves is remotely smooth or flat. This
REVERBERANT SOUND
tells us that all of the sounds arriving at the ears do not convey the
20 50 100 500 1K 5K 10K 20K
FREQUENCY (Hz) same message about the sound quality, or timbre. The top curve is a
26
calculated prediction of what a measured “room curve” might be.
The loudspeaker was then placed in a typical left or right channel
location in a real room. It was measured at the listening position,
Real vs. Predicted Room Curves
30 then moved to two other locations within a radius of two feet, and
measured at each location. The fourth curve, the top one, is the
20 calculated room curve from the previous slide. Obviously, little is
Three “typical” locations
dB
10
close to a room corner. changing at frequencies above about 300-400 Hz, and the prediction
is right on target. However, below these frequencies, there are
0 considerable location-dependent changes, and the prediction fails
-10
completely. The reason? Room resonances and boundary effects
20 50 100 500 1K 5K 10K 20K that are specific to that particular room. These can only, with
FREQUENCY (Hz) 27
precision, be evaluated by measurements in the room itself.
There is a region where the room dominates,

and one where the loudspeaker dominates.
However, it is also clear that through the middle and high frequencies
30 the anechoic measurements made in the laboratory have done an
excellent job of predicting what happened in the room. However,
20
ROOM SPEAKER
doing so required many, many measurements at positions all around
dB
10 the loudspeaker.
0 If we are to try to anticipate how a loudspeaker will sound in a room,

-10
it is necessary to measure everything, and not just a few curves
20 50 100 500 1K 5K 10K 20K around the on-axis measurement.
FREQUENCY (Hz) 28
31 January, 2002 6
At Harman, the engineers have dubbed our basic measurement of

frequency response, the Spin-o-rama, since it involves spinning the
loudspeaker on two axes, and accumulating a total of 72
measurements.
29
The collection of raw data is computer processed to generate a set of

curves showing estimates of the distinctive regimes of sound arriving
at a listener’s ears in a typical room. To do this, a large survey of real
rooms was undertaken, and a statistical analysis of angles and
distances led to the algorithm that generated these curves. All
DIRECT SOUND measurements have a frequency resolution of 1/20 octave, from 20
Hz to 20 kHz.
The top curve is the on-axis curve, representing the direct sound for a
person in the “sweet spot”.
30
The second curve is a spatial average over +/ - 30° horizontal, and +/-
10° vertical, representing the direct sound for listeners seated in a row
LISTENING of chairs or a large sofa, and possibly standing and sitting.
WINDOW
31
The third curve is the energy sum of the set of early reflections.
EARLY
REFLECTIONS Ideally, these should look a lot like the on-axis curve, so that it
conveys the same timbral information.
32
The fourth curve is a calculation of the total sound power radiated by

the loudspeaker in all directions (this is NOT a simple average or sum
SOUND POWER of all 72 measurements). Again, this curve should be smooth and
flattish.
33
31 January, 2002 7
The uppermost of the bottom pair of curves is the Directivity Index,

or DI. This is an indication of the angular uniformity with which the
loudspeaker radiates its energy into a room as a function of
frequency. It is a measure of the uniformity of its dispersion as a
function of frequency.
DIRECTIVITY
INDEX
34
The bottom curve is an invented DI, this time just for the early
reflections.
EARLY
REFLECTIONS
DIRECTIVITY
INDEX
35
Here we see the complete set as they are presented for visual
The Complete Set – inspection. The whole idea of this is to present to the eyes, a set of
A Picture of Performance data that can be interpreted in a way that allows one to anticipate how
a loudspeaker might sound in a room.
The curves shown here describe a truly excellent loudspeaker, not
perfect but, currently, a good example of the state of our art. Note the
smoothness of all of the curves, and the basic similarities in all of the
10 dB
5 dB DI curves, from the single on-axis measurement, through to the estimate
0 dB
of the total sound radiated in all directions, the sound power.
36
Performance in an “average” room

can also be predicted
1/20-octave resolution.
Measured average of 4 different rooms
In the same way that the earlier example was calculated, we can
generate a curve that tries to anticipate a room curve. Here we have
not included all of the possible refinements, but it suffices to get us
Predicted from anechoic data
within a couple of dB over most of the frequency range.
37
Performance in an “average” room

can also be predicted
1/20-octave resolution.
125 Hz to 8 kHz
8-octaves
with useful accuracy
38
31 January, 2002 8
Now, how do we interpret the

measurements?
• Frequency response curves are not flat and
smooth. Does this matter? How much does O.K. So we get some curves. The real proble m is that they ARE
it matter? What is the ideal shape?
curves, and not straight lines. What is the ideal shape? How much
• Can we hear phase shift?
• What about time-domain behavior: transient
deviation from the ideal is audible? Is there more to this than just
response, “speed”, etc.? frequency response?
39
Let’s start with the most basic of all measurements, the frequency
For Example: What is in a response. In the case of a loudspeaker we would begin with a look at
Frequency Response? what happens on the major axis.
Incidentally, such a measurement should be made at a distance of 2 m
(6 feet) or more. The industry standard specifies loudspeaker
sensitivity at one meter, however, the standard also requires the
measurement to be made in the “far field” of the source, and if
necessary, for the measurement to be calculated back to 1 meter.
0 Degrees / On Axis Many people mistakenly do not do this, and also make frequency
response measurements at 1m. For loudspeaker systems of typical
40
size, these measurements can exhibit large errors.

Of the features that our eyes can extract from a curve like this, it is
obvious that spectral balance and bandwidth are important. Try
What is in a Frequency Response?
playing with the bass and treble controls, and you will find that small
Spectral Balance
changes are audible.
and Bandwidth Resonances are REALLY important because our perceptual system
(the ears and brain) is highly sensitized to them. The reason:
Resonances
resonances are the “building blocks” of all of the sounds that we are
really interested in listening to – voices and musical instruments.
Acoustical Resonances can cause peaks and dips in a frequency response curve.
Interference
However, so can acoustical interference, a phenomenon that turns out
41
to be much less audible under normal listening circumstances.
The components
must be separated
Spectral Balance
So, we need a measurement system that allows us to separate,
and Bandwidth visually, those features in a curve that are caused by each of these
phenomena. Only then can we be truly analytical, and make good
Resonances
judgments about how good or bad the device is.
Acoustical
Interference
42
A Traditional, On-Axis
Frequency Response
Measurement
Once upon a time, it was thought that a single curve told us useful
information.
0 Degrees / On Axis
43
31 January, 2002 9
Simple Spatial Averaging

Then we learned that spatial averaging allows us to separate those
10º UP
peaks and dips caused by resonances from those caused by acoustical
30º R interference. The explanation is really simple: those features
20º R
associated with resonances tend not to change when the microphone
0º 10º R
30º L
20º L 10º L location is changed, while those associated with interference do.
THE LISTENING
WINDOW
10º DN
44
A lot of Spatial Data

When we average a lot of measurements made at a lot of different
locations, and certain visual shapes do not disappear, we can be quite
confident in concluding that those are resonances, and not the result
of acoustical interference.
45
Spatial Averaging in Rooms

In rooms, there is an abundance of acoustical interference, caused by
• Helps to reveal the presence of resonances, multitudes of reflections. Therefore, spatial averaging, i.e.
which can be equalized. combinations of measurements made at several locations, can help to
• Attenuates the effects of acoustical
interference, which cannot be equalized. isolate resonances. This is important because it turns out that we can
• Helps to eliminate visual evidence of dips in equalize resonances (about which, more later), and we cannot
the frequency response, and thereby the equalize the effects of acoustical interference.
temptation to try to fill them with
equalization – which does not work!
46
Resonances are major problems!

• In loudspeaker drivers So, we expend a great deal of effort eliminating resonances from
– cone flexure modes loudspeaker systems, and when they are installed in rooms, we need
– suspension and frame modes
to spend some time and effort to identify and eliminate serious
• In enclosures
– mechanical resonances in panels and surfaces
resonance problems.
– acoustical resonances in cavities
• In rooms
47
Resonances are differentiated by their “Q”, or quality factor. A high-

Q resonance is one that is very frequency specific and that rings a
Resonances of Different “Q”
long time. An example of a high-Q resonance is an empty wine glass,
Q=1
held by the stem, and tapped with a finger nail. It emits a clear tone
10 dB
that rings. If one places a finger on the side of the glass and taps it
again, the ringing is shorter. The finger has taken some energy out of
Q = 50 Q = 10 the resonant system, and the ‘quality’ is reduced. If the entire glass is
grasped by a hand, and the tap is repeated, there is almost no ringing
100 200 500 1K 2K 5K 10K
at all. A tone is still recognizable, but it is a low ‘quality’, or low-Q,
FREQUENCY (Hz)
resonance. High-Q resonances have sharp peaks, and low-Q
48
resonances are broader when they are seen in frequency responses.
31 January, 2002 10
The just audible variations in spectrum Sean Olive and I, when we were at the National Research Council, in
or frequency response:
Canada, published a paper in which we showed the shapes of
High-Q resonance Q = 50
20
deviations in frequency responses that corresponded to the just
PROGRAM: VOCALIST & JAZZ COMBO
SOUND LEVEL (dB)
10 10 dB
audible thresholds for resonances of different Q, at different
0
frequencies, for different kinds of music or sounds. The effect of
frequency was secondary, so here I show only what happens at 500
-10
Hz. The results at different frequencies are similar. It shows that, for
-20
20 50 100 200 500 1K
FREQUENCY (Hz)
2K 5K 10K 20K multimiked pan-potted, low reverb, pop or jazz, the threshold of
audibility corresponds to a 10 dB spike in a frequency response
49
curve. It looks bad, but it is just barely audible!
The just audible variations in spectrum

20
With a big band or symphony orchestra (complex orchestration) in a
PROGRAM: SYMPHONYORCHESTRA
PROGRAM:SYMPHONY ORCHESTRA
SOUND LEVEL (dB)
10
reverberant hall, the threshold is lower (we are more sensitive).
6 dB
0
-10
-20
20 50 100 200 500 1K 2K 5K 10K 20K
FREQUENCY (Hz)
50

20
Of all the signals we tested, pink noise was the most revealing of
PROGRAM: PINK NOISE
SOUND LEVEL (dB)
10
resonances. It produced the lowest thresholds. Such low-amplitude,
3 dB narrow, spikes are difficult to measure with precision at all
0
frequencies.
-10
-20
20 50 100 200 500 1K 2K 5K 10K 20K
FREQUENCY (Hz)
51

Medium-Q resonance Q = 10
20
SOUND LEVEL (dB)
10
When the Q is reduced, the pattern of audibility is much the same, but
0
6 dB the thresholds are lower.
-10
-20
20 50 100 200 500 1K 2K 5K 10K 20K
FREQUENCY (Hz)
52

20
PROGRAM: SYMPHONY ORCHESTRA
SOUND LEVEL (dB)
10
4 dB
0
-10
-20
20 50 100 200 500 1K 2K 5K 10K 20K
FREQUENCY (Hz)
53
31 January, 2002 11

20
PROGRAM: PINK NOISE
SOUND LEVEL (dB)
10
0 1.5 dB
-10
-20
20 50 100 200 500 1K 2K 5K 10K 20K
FREQUENCY (Hz)
54
The just audible variations in spectrum When we get to really low-Q resonances, the ones that ring very little,
it turns out that we can hear them at very low measured amplitudes.
Low-Q resonance Q = 1
20
What, then, of the arguments that the ringing of high-Q resonances
SOUND LEVEL (dB)
10
“smears” sounds, making them less articulate? These are arguments
0
2.5 dB that are most likely based on visual interpretations of measured data,
not on actual subjective tests of the audibility of the effects. They
-10
-20
sound as though they should be true but, except at very low
20 50 100 200 500 1K
FREQUENCY (Hz)
2K 5K 10K 20K frequencies, they are fanciful. Good engineering should attempt to
eliminate resonances of all kinds, but it is important to understand
55
what is and is not audible.

20
SOUND LEVEL (dB)
10
0 1.5 dB
-10
-20
20 50 100 200 500 1K 2K 5K 10K 20K
FREQUENCY (Hz)
56

This “curve” looks almost like a straight line. Our eyes are telling us
20
that it is almost perfect, yet our ears are telling us that there might still
PROGRAM: PINK NOISE
SOUND LEVEL (dB)
10
be something audibly wrong.
0 0.3 dB
So, in this case, what our eyes tell us does not intuitively correspond
with what we hear. This is why it is so important to do the science,
-10
and to establish what the real psychoacoustic relations are. Our
-20
20 50 100 200 500 1K
FREQUENCY (Hz)
2K 5K 10K 20K instincts can be wrong.
57
Is there an explanation? It is probably because music and speech are

The just audible variations in spectrum ever-changing. Also, voices and many musical instruments are
played with vibrato – a modulated pitch. High-Q resonances take
These are all EQUALLY audible!
20
time to build up, as well as to decay. We tend to talk about the
SOUND LEVEL (dB)
10
ringing, overhang or decay of resonances after the signal has stopped,
0
ignoring the front-end effect. High-Q resonances are narrow, very
frequency-specific, and musical sounds must be sustained long
-10
-20
enough to energize them. Few are. Low-Q resonances are wide
20 50 100 200 500 1K
FREQUENCY (Hz)
2K 5K 10K 20K enough that they respond to everything, and they take almost no time
to reach full amplitude.
58
31 January, 2002 12
In order to make any sense at all of a frequency response curve, or a

set of curves, they must be capable of revealing to our eyes
everything that is audible.
Measurements must have enough
resolution to show what we can The belief, still widespread in this industry, that we cannot measure
hear what we can hear, has its origins in situations where the measured
data were erroneous or incomplete. Such situations are common in
the loudspeaker business.
59
A Resolution Test
Let’s create a test. Suppose we had an imaginary system in which
10
there were high-Q resonances uniformly distributed from low to high
dB frequencies. A competent measurement system would reveal them to
0 our eyes as they truly are.
20 50 100 500 1K 5K 10K 20K
FREQUENCY (Hz)
Q=50 resonances at the threshold of audibility
60
However, not all measurement systems are equal. Many very

commonly used ones “gild the lily”, making the curves smoother than
A Partially “Blind” Measurement
they really are. All systems that use time windowing, or the
10
True Level equivalent (MLSSA, TEF, and any FFT-based system) can do this IF
dB
the measurement window is not sufficiently long. Here I show what
0 happens with a quite long window (17ms), more than is used by many
CANNOT MEASURE WHAT WE HEAR manufacturers and reviewers. It is clear that the measurement does
20 50 100 500 1K 5K 10K 20K
FREQUENCY (Hz) not reveal the existence of the high-Q resonances in the middle and
Measurement of Q=50 resonances at the threshold of low frequency regions. It does not show things that we know we can
audibility using a time window of 17 ms (60Hz)
hear.
61
A Totally “Blind” Measurement

The popular one-third-octave measurements, very common in room
10
True Level measurements, simply fail. These give only a very “broad-brush”
dB
view of what is happening, and are of very limited use. One needs an
0 analyzer capable of at least 1/10 octave resolution in order to reveal
CANNOT MEASURE WHAT WE HEAR what we need to see.
20 50 100 500 1K 5K 10K 20K
FREQUENCY (Hz)
Measurement of Q=50 resonances at the threshold of
audibility using a 1/3-octave analyzer
62
For loudspeaker measurements, very long measurement windows are

necessary, and these can be accomplished only in anechoic spaces.
A 4 π (full space)
anechoic chamber Outdoors, away from all reflecting surfaces, is free, but impractical.
Anechoic chambers, such as the one shown here, are very practical,
4-foot long wedges:
anechoic to 60Hz, but also very expensive. However, this is the price of entry if you are
calibrated to 20 Hz
seriously in the business. The length of the wedges determines how
Automatic positioning:
motorized, computer-
low in frequency one can measure accurately. These 4-foot wedges
controlled platform create a reflection-free environment down to about 60 Hz. We have
calibrated it down to 20 Hz for specific measurement locations within
it. With a large enough measurement time window, ANY
63
measurement system should yield accurate data.
31 January, 2002 13
AND the most commonly used

specification for frequency
response is useless . . .
. . unless it is accompanied by
a graph!!!
64
20 Hz to 20 kHz +/- 3 dB
+3
dB 0
-3 Every audio device has a specification for frequency response. A
+3
dB 0
tolerance of +/- 3 dB is sufficient to describe a range from junk to
-3 jewels. By itself, it is meaningless window-dressing. A curve, and
+3
dB 0
the ability to interpret it, are necessary. If the tolerance is small
-3 enough, then it does have meaning, of course.
+3
dB 0
-3
20 50 100 200 500 1K 2K 5K 10K 20K65(Hz)
Frequency and Time Domains So far, we have talked about frequency response as though it were the
only important factor. What about the all-important transient
• Related by Fourier Transformation response, speed, punch, drive, and all of those descriptors of what
• Behavior in one domain is paralleled by happens in the time domain?
corresponding behavior in the other
Well, it turns out that the two domains are related to one another, by
the Fourier transformation.
66
A perfect linear system would be described either by a clean

uncluttered transient, or by a pair of flat straight lines portraying a
Frequency and Time Domains
constant amplitude vs. frequency characteristic (we call this the
frequency response, although it is really the amplitude response), and
A
a constant phase vs. frequency response. The flat amplitude response
A TIME FREQUENCY tells us that the signal level at all frequencies is constant. The flat
ø phase response tells us that everything is happening at precisely the
right instant in time. The combination of flat amplitude and phas e
A PERFECT SYSTEM FREQUENCY responses correspond to a perfect impulse, or transient, response.
67
Here we have disrupted the perfect system with a single high-Q

resonance. The narrow “footprint” in the amplitude response, as seen
earlier, is repeated in the phase response. In the time domain, the
4 dB corresponding effect is extended ringing (the empty wine glass).
A
A TIME FREQUENCY If we measured the amplitude and phase responses, a computer could
ø perform a Fourier transform and give us the transient response. If we
AN OTHERWISE PERFECT
measured the transient response the computer could calculate the
SYSTEM WITH A Q=50 FREQUENCY amplitude and phase responses. So the information on the left side of
RESONANCE AT THE
THRESHOLD OF AUDIBILITY the slide is the same as that on the right side, only displayed in a
68
different form.
31 January, 2002 14

2 dB
A
Here is a medium-Q resonance. The frequency-domain ‘footprint’ is
A TIME FREQUENCY larger, and the time-domain ‘footprint’ is smaller.
ø
SYSTEM WITH A Q=10 FREQUENCY
RESONANCE AT THE
THRESHOLD OF AUDIBILITY
69

0.5 dB
A
And a low-Q resonance. Note the convenient relationship: as the
A TIME FREQUENCY footprint in the frequency domain gets larger, that in the time domain
ø gets smaller.
SYSTEM WITH A Q=1 FREQUENCY
RESONANCE AT THE
THRESHOLD OF AUDIBILITY
70
There is a class of systems that behave as “minimum-phase” systems.

In such systems, if one has measured only the amplitude response –
“Minimum-Phase” Systems our familiar “frequency response”, it is possible to calculate the phase
response from that data. Now, if we know both the amplitude and
A TIME A
phase responses, we can calculate the time response. So, in a
FREQUENCY minimum phase system, a measurement of the frequency response,
allows us to predict the time response. A bump in the frequency
IN “MINIMUM-PHASE”
SYSTEMS THE PHASE response means that the system must ring. A flat, smooth, frequency
RESPONSE CAN BE ø response means that there is no ringing. The previous data show that
CALCULATED FROM
THE AMPLITUDE RESPONSE.
FREQUENCY we are able to measure the visual evidence of audible resonances in
71
frequency response curves. This is really important.
What Systems Exhibit Minimum

Phase Behavior?
• Many common functions in analog and Several very important devices are minimum phase systems, meaning
digital electronics. that, for these devices, the frequency response curve is the single
• Loudspeaker transducers – conventional
woofers, midranges and tweeters. most important measure of audible performance in the linear domain.
• Room resonances at low frequencies. Of course we do measurements of non-linear effects as well, but in
general these are much less troublesome.
This is a HUGE advantage!
72
A minimum-phase system with a

problem resonance.
A
A TIME FREQUENCY If a minimum phase system has a resonance, and we wish to get rid of
ø the audible effects, we can choose to do it electronically.
FREQUENCY
73
31 January, 2002 15
Address the resonance with an equal Simply design a minimum-phase filter, in either analog or digital
and opposite parametric EQ filter electronics, that exactly matches the shape of the bump in the
frequency response, but is inverted. When the two are added, we get
A
a straight line. The filter, because it is minimum phase, will have a
WHEN THE CORRECT
AMPLITUDE RESPONSE FREQUENCY phase shift that mirrors the phase shift in the resonance, so that a
IS “DIALED IN”, THE
PHASE RESPONSE IS ø summation yields another straight line.
AUTOMATICALLY
CORRECTED.
FREQUENCY How often have you heard that equalizers are bad because they add
phase shift? Here we show that it is a good thing – assuming that it
74
has all been done properly, with the necessary precision.
Two flat lines on the right, we know, correspond to a perfect impulse

And everything is fixed!
response on the left.
A
This is a very simple form of pre -distortion, a well known technique
A TIME FREQUENCY that, with the advent of digital processing, is likely to become more
ø widespread. If we know what an electromechanical or acoustical
system is doing wrong, there are some things that can be corrected by
FREQUENCY modifying, or pre-distorting, the signal so that what is eventually
radiated as sound is correct.
75
Which is one reason why It is not magic, but it certainly seems like it. Good loudspeakers can
active/amplified loudspeakers are be made better. Room resonances can be tamed (for specific listeners
attractive. at least).
A good loudspeaker without However, in order for it to work we need accurate, high resolution,
equalization can be an even better frequency response data, and parametric filters.
one with the right kind of
equalization.
76
Spatial Averaging ADDS Information

Which brings us back to measurements and spatial averaging. Here
LISTENING WINDOW ± 30º HOR. ± 10º VERT.
we see a portion of a spin-o-rama for a loudspeaker showing some
ON AXIS
bumps in the on-axis curve. The bumps are attenuated, and even
disappear, with spatial averaging. This tells us that the bumps are
THESE ARE NOT
RESONANCES caused by acoustical interference (in this case diffraction from the
AND THEY SHOULD cabinet edges). They are not resonances, and they should not be
NOT BE EQUALIZED
SOUND POWER equalized.
77
In contrast – these are resonances, In this example, the series of bumps penetrate all of the curves,
and they can be equalized surviving even the 72-curve calculation of sound power. These truly
are resonances, and they can be treated with individually designed
parametric filters.
Spatial averaging ADDS information. Spectral averaging

(smoothing) takes it away.
78
31 January, 2002 16
Measurements make a nice story,

but
can people really hear the If all of this really means anything, we should be able to prove it
differences? using our carefully conducted listening tests.
Let’s test them with four

“high-end” speakers.
79
Four expensive and highly regarded loudspeakers are evaluated in the

“shuffler” room, in a double-blind test. These are the measurements.
This one looks good: $10,000/pr
The listeners, of course, do not get to see them until it is all over.
This one is well behaved. It has smooth, flattish curves, very wide,
very uniform dispersion, and excellent low-frequency extension. The
slight sag in the upper middle frequencies is something that is
sometimes done to compensate for the numerous excessively bright
recordings out there. The trade-off, others might sound a bit laid
back.
80
This one too: $8000/pr

Here is a ‘tell it the way it is’ speaker. Unabashedly flat, very
smooth, superb low bass extension, being only about 5 dB down at 20
Hz, with a directivity that smoothly and gradually rises with
frequency. A small dip in sound power around 2 kHz might be barely
audible, but likely only in quite ‘live’ rooms.
81
This one is likely to have a ‘personality’. The undulations in the

upper mids/lower treble are everywhere, including the directivity. It
A bit wobbly: $8000/pr
is possible to play detective, and guess that this is a three-way system,
with the woofer crossing over to the midrange around 300 Hz, and the
midrange crossing over to the tweeter around 3-4 kHz. How do we
know? Look at the directivity curves. Very low frequencies are
omnidirectional. The curve rises as the woofer becomes more
directional until it crosses over to the smaller midrange when the
directivity drops. It then rises again with frequency until it crosses
over to the small tweeter, at which point the cycle begins again. The
82
low bass is fair, but a small bump just below 100 Hz spoils it.
This speaker has a bunch of things going on. Clearly the designers
didn’t believe that flat was necessary, or they didn’t know how to
A Mountain Skyline: $5000/pr
achieve it. Not only are the general trends not flat, but superimposed
are peaks and dips suggesting resonances. The proof that they are
resonances is in the fact that the patterns are repeated in all of the
curves. The directivity is interesting, being zero up to 100 Hz (the
woofer) and then abruptly rising to about 5 dB and hovering around
that all the way to 20 kHz. Since 4.8 dB is the directivity of a dipole,
we could suspect that this is a hybrid system with a panel loudspeaker
operating above about 100 Hz. The woofer exhibits a significant
83
bump and then rolls off below about 60 Hz. No subwoofing here.
31 January, 2002 17
After several rounds of listening to different kinds of music, several

listeners yielded subjective preference ratings that were processed in
The Listeners Speak
a statistical analysis program. One of the results is a bar graph
Cell Bar Chart
showing the average rating for the group of listeners, for each of the
Grouping Variable(s): Loudspeaker
8
Error Bars: 95% Confidence Interval loudspeakers. The tiny lines on top of the bars show the 95%
7
6
confidence intervals. If the differences in the ratings are greater than
5 these lines, the differences are probably statistically significant, and
4
3
not due to chance. The two top-rated speakers are not significantly
2
1
different from each other, according to this rule. The other two are
0
Brand Y Brand X Brand B Brand A
truly less good.
84
When we combine the subjective with the objective data, it is clear

that the loudspeakers that yielded the best set of technical data, also
were preferred by the listeners.
Cell Bar Chart
Grouping Variable(s): Loudspeaker
Error Bars: 95% Confidence Interval
It works!
Brand Y Brand X Brand B Brand A 85
At Harman, we do hundreds of such listening evaluations, using

What do listeners say they like?
competitors’ products that we purchase on the open market. It is
$2000/pr essential to know where our new products stand with respect to the
competition.
The results are monotonously the same. Loudspeakers that look good
in the spin-o-rama measurements are the ones that are subjectively
preferred.
86

This is one of those pricey ‘high end’ bookshelf-sized speakers that
$8000/pr some reviewers have raved about. The measurements suggest that it
is a slightly dull sounding, moderately colored system with no real
bass. The listeners agreed.
87

$10,000/pr Note that price sometimes has nothing to do with sound quality.
There may be material value to justify the high price, a sexy
appearance, or just a lot of hype. This is the audio business, and
reason and the laws of physics seem not to be universally applicable.
88
31 January, 2002 18

$1800/pr
Truly excellent sound can be found at moderate prices.
89

$1800/pr
90

$700/pr
91

$900/pr
92

$300/pr Even entry-level products can sound good. They will come in plain
enclosures, and they may not play as loud as the bigger boys, but they
are eminently capable of preserving most of the artistic integrity.
93
31 January, 2002 19

$460/pr Here’s one to avoid! It is amazing that anyone, especially a well-
advertised brand as this one is, would actually let something like this
out into the marketplace. For the same price, they could be selling
good sound. Obviously they don’t care. Instead they sell a slick
package and a story.
94
ALL of the most preferred loudspeakers are ones that exhibit the
flattest, smoothest families of curves.
They exhibit the fewest, and the lowest level, resonances. They have
Conclusion: the flattest, smoothest, widest bandwidth frequency responses when
measured from all angles.
Listeners don’t like resonances!! They have similar shapes in all of the curves – i.e. they have quite
constant, or at least smoothly changing, directivity as a function of
frequency.
Can we measure what we can hear? No, but we sure have made a
good start.
95
This is a powerful position to be in, when it is possible to demonstrate

Equivalent Measurements
that the right set of accurate measurements has a consistent
relationship with listener evaluations.
We do not claim to have mastered everything, at this stage. However,

SUBJECTIVE OBJECTIVE
some things are understood. They even make logical sense.
So, let’s assume that, to a first approximation, we understand how to

design loudspeakers that have the potential of sounding good in a
96
room.
The Rules for Good Sound in

Rooms
At middle and high frequencies:
ü Start with a loudspeaker that was designed to The second rule for good sound requires that we look at some
function well in a variety of different rooms.
Ø Use geometry, reflection, diffusion, and absorption to specifics of the room itself.
achieve good imaging and ambiance.
At low frequencies:
Ø Maximize the output from the subwoofer(s).
Ø Achieve a uniform performance over the listening
area.
Ø Equalize to achieve good performance.
97
Reflections alter both

Sound Quality and Imaging
Reflected sounds can be controlled by:
(a) controlled-directivity loudspeakers,
(b) absorbing or diffusing objects on
reflecting surfaces in the room,
(c) the shape of the room,
(d) some of each.
98
31 January, 2002 20
Here is a cartoon description of what happens with reflected sounds

in a room.
We start with only a floor, no walls. Brunhilde, of opera fame, is

singing in the right speaker only.
BRUNHILDE IN THE RIGHT CHANNEL

99
Adding walls, the one next to the right loudspeaker is some distance
away, produces a nice warm “spatial” illusion. It sounds a little
richer.
ADD WALLS, THE RIGHT ONE FAR AWAY

& WE HEAR A ROOM AROUND THE IMAGE 100
If the wall is moved a bit closer to the speaker, we note that the lady
is a little bit “smeared”, putting on some weight, and maybe leaning a
bit to the right.
MOVE IT CLOSER AND THE FAT

LADY SINGS 101
If the wall is too close, the truly fat lady is singing.
STILL CLOSER AND THE REALLY

FAT LADY SINGS 102
Why? Because the wall is an acoustic mirror, creating a second

acoustical loudspeaker, just as it would create a second visual one if
the wall were optically reflective. No wonder things got a bit fuzzy.
THE REASON:
THE WALL IS AN ACOUSTIC MIRROR,
CREATING A DUPLICATE SPEAKER
103
31 January, 2002 21
Placing some sound absorbing material on the wall, at the reflection

point (have a helper hold a mirror against the wall and find the
location where you can see the loudspeaker tweeter from the most
important listening position). The reflected sound is attenuated, and
the lady loses a bunch of weight.
What material? Acoustic foam or rigid fiberglass board, with or
without acoustically transparent fabric covering.
How thick? Not less than one inch, preferably two to four inches.
SOUND ABSORBING MATERIAL ON THE WALL
REDUCES THE “SIZE” OF THE REFLECTION
How large? To be really effective, a patch at least 3 to 4 feet on a
AND THE LADY IS SLIM AGAIN side is necessary. Tiny little “cushions” are more psychological than
104
acoustical. Heavy, velour drapes, densely folded also work well.
NARROW DISPERSION WIDE DISPERSION

FOCUSES SOUND ON
LISTENERS, MINIMIZING
“ILLUMINATES’ THE
ROOM, CREATING MANY
Nowadays, we know enough about horn design to be able to make
WALL REFLECTIONS,
MAXIMIZING CLARITY
REFLECTIONS WHICH ADD
SPACIOUSNESS, EVEN IN
them sound really good, and take advantage of their directional
AND INTIMACY
AS INTENDED IN MOVIES.
SCENES WHERE IT IS NOT
APPROPRIATE.
control. The days of horns that are just loud and sound like
megaphones are past – for good engineers at least.
If the room is acoustically live (the way many interior decorators like
them), then the only option is to use horns, or waveguides, to control
the radiation from the loudspeakers. This way the energy is focused
on the listeners, and kept away from the reflecting surfaces,
improving the intelligibility and directional effects.
105
The create and destroy method Movies, especially, are designed for listeners in a strong direct sound
ADDING SOUND ABSORBING field. Some people use wide-dispersion loudspeakers, and then cover
MATERIAL ALONG THE SIDE
WALLS CAN the walls with sound absorbing material. This gets the job done, but
CREATE MUCH THE SAME
EFFECT AS DIRECTIONAL in doing so it makes the entire system work harder, first to create the
LOUDSPEAKERS.
HOWEVER, IT IS VERY
sound, and then turning it into heat in absorbers. The result, dynamic
WASTEFUL
TO CREATE THE SOUND
range is sacrificed. Not necessarily a good tradeoff.
ENERGY Acoustically dead rooms are also not very pleasant places in which to
AND THEN TURN IT INTO
HEAT spend time, conversing or anything else. Some custom home theaters
IN AN ABSORBER!
are like this. It is not a recommended solution.
106
A Basic Imaging Test Whether it is a classic two-channel stereo system, or a multichannel

THE SIGNAL: BROADBAND MONOPHONIC PINK NOISE system, one of my first tests is to play monophonic pink noise
(available on numerous test CD’s) through the front left and right
speakers, sit in the “sweet spot” and listen. What should be heard is a
compact image of noise, floating midway between the loudspeakers.
As you move backwards in the room the image should stay. As you
lean left or right, the image should move left or right. This is normal.
It is a phantom stereo image.
107
“Normal” Stereo Imaging

Now, put on some music. The featured artist in pop and jazz
recordings should float in the middle location. The band should be
across the front creating a solid sound stage (the success of this is
greatly dependent on the recording, so be sure to try a few). In
recordings with “ambiance”, like most in the classical repertoire, you
might sense an acoustical spaciousness around you. This is good.
108
31 January, 2002 22
“Tightening” the Image
Some two-channel customers like to “get into the image”. For them
you can suggest some absorbing material, even heavy drapes will do,
along the side walls. This attenuates the side wall reflections and the
image “tightens” up nicely. Moving the curtains away, opens up the
‘space’ again.
109
Opening it up with Reflections
Other customers like to think that they are in the concert hall. For
them room reflections are not necessarily a bad thing. In fact, you
might consider adding a few more, using some of the commercial
diffusing elements on the market. Just be careful not to overdo it.
The test is that the center image stays intact even when you move to
the rear listening locations.
110
Even good things can be taken too far. I have been in recording
"Confusers" and Confusion control rooms where so much diffusion has been added that the center
image is completely destroyed! The noise ‘image’ was the entire
front wall. And recordings were being made in this situation! This
design was fashionable – yes there are ‘fashions’ in acoustics too – a
few years ago. Just as in many things, some fashions are just silly.
This one was aided by the other fashion of that period: the live-
end/dead-end room, another case of an idea taken to excess. It helped
Too much of anything can be bad
some bad studio monitor speakers sound better, but it is not
something to be recommended, certainly not for recreational
111
listening, and not for multichannel sound.
"Confusers" and Confusion
This becomes
Many homes do not allow us the luxury of sitting away from the back
especially wall. In those cases the last thing one would do is put diffusers
important if the directly behind the listeners’ heads. Even a hard flat wall can disrupt
back wall is close
to the listener(s). the front soundstage. A simple demonstration can convince you, or
your customer that something is wrong. While listening to the mono
pink noise, just hold an upholstered cushion or pillow behind the head
of a listener in the stereo seat. Usually the image tightens right up.
112
Designing the Listening Experience
This becomes
especially
important if the
back wall is close
A patch of absorbing material is a much better solution. Use diffusers
to the listener(s). on the sides, if you like.
In such cases,
some absorption
is appropriate.
113
31 January, 2002 23
What is a “Diffusor”?
Commercial diffusers are highly specialized devices, designed to

It can be a special accept sounds arriving from any angle, and then to re-radiate them in
surface designed all directions. Such diffusers, then, need to be considered as
to “reflect” sounds that
arrive from any direction, distributed sound sources.
off in all directions.
It becomes a “distributed source”
with all parts sending some sound
back towards the listener.
114
To a listener, these surfaces send a large number of individual

reflections to the ears, from all parts of the device.
115
The classic “polycylindrical” diffuser, is nothing more than a curved

surface intended to break up large flat surfaces. As diffusers they
Or, it can be a shaped surface
work very well indeed, and they are inexpensive. They can also be
incorporated into interesting looking architectural features, possibly
including lighting effects. If you want to get creative, there are many
regular and irregular geometrical shapes that work well. A good dry-
wall artisan will love you for giving him something interesting to do.
Here, the conventional
laws of reflection are Remember to bounce some of the sound vertically too. If the
obeyed diffusion is to be effective over middle as well as high frequencies,
some of the shapes must be a foot or more deep. The notion that
116
textured paint does anything consequential is another fantasy.
A listener receives only one reflection from each of the curved

surfaces.
117
Or it can be furnishings: bookcases,

cabinets, fireplaces, etc.
If the listening room is also a normal living space, it may not be
necessary to use any special acoustical devices at all. With a little
thought, bookcases, display cases, paintings, fireplaces, etc. can all do
the job without making the room look at all “technical”.
118
31 January, 2002 24
In contrast, a wall . . .
Flat empty walls not only look stark, but they sound that way too.
119
120
An acoustical consultant walks into a room, stands by the door, claps

his hands, furrows his brow, and pronounces that this room has really
Evaluating Flutter Echoes
bad flutter echoes and you need his (expensive) help to fix it. If this
happens, say goodbye.
The only flutter echoes that are important to the quality of reproduced
sounds in the room are those that are excited by the loudspeakers
themselves. Have an assistant clap hands at the loudspeaker locations
while you listen from the relevant locations in the room. If there is a
problem then fix it. It matters not that flutters can be heard from the
top of a step ladder.
121
Geometric irregularities on walls, It is amazing how little it takes to cause an audible flutter, and it is
furniture and diffusing elements amazing how little it takes to get rid of one. I have seen a picture,
are excellent cures for flutter hung on a slight angle, do the job. Moving a bookcase, adding a wall
echoes. bulge over a fireplace, a two-foot square patch of diffuser or absorber
They eliminate the problem in a large wall, all have solved annoying problems without absorbing
without absorbing sound. significant sound.
122
Coming up in Part 3
Perfecting the low frequencies. How many subwoofers? Where do I put them?
Where do I sit? What about “bass traps”? How do I get rid of “room boom”?
Here we look at what it needed for truly excellent bass performance in rooms.
An understanding of room modes, or resonances, is essential to achieving
uniform bass over a listening area. The right kind of equalization can help to
31 January, 2002 25
make that bass sound good, but it cannot do everything. Some traditional
forms of equalization have a good chance of getting it wrong. Interestingly,
two or more subwoofers, strategically located, can be very beneficial.
31 January, 2002 26

Harman / Kardon: Loudspeakers and Rooms For Multichannel Audio Reproduction

Uploaded by

Copyright:

Available Formats

Harman / Kardon: Loudspeakers and Rooms For Multichannel Audio Reproduction

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Harman / Kardon: Loudspeakers and Rooms For Multichannel Audio Reproduction

Uploaded by

Copyright:

Available Formats

The Science of Audio - a series of lectures by Floyd E. Toole, Ph.D.

Vice President Acoustical Engineering

Part 2 – Making a good loudspeaker -

Loudspeakers can be designed to be “room

Some of what we think we

The fact is that audio is a technology with a firm scientific grounding

The existence of custom audio installers and consultants is something

his or her own home, to achieve the best possible sound.

They affect sound quality and imaging

When we listen we are instantly trapped in the “audio circle of

boxes, mini-systems, headphones and entry level car audio? It isn’t!!

The Audio Industry is in a “Circle of Confusion”

Ironically, the problem exists at both the professional and consumer

If listeners like or dislike something, it is important to try to identify

the audio industry find these double-blind tests very threatening.

locations when they are being auditioned. It is pneumatically driven,

A pair of speakers ready for listening.

And, about three seconds later, a second pair can be heard.

The Room - Direct Sound

The first sound to arrive at a listener’s ears is the “direct” sound. If

The Room - Early Reflections

The Room – Resonant Modes

There is a region where the room dominates,

0 If we are to try to anticipate how a loudspeaker will sound in a room,

At Harman, the engineers have dubbed our basic measurement of

The collection of raw data is computer processed to generate a set of

The fourth curve is a calculation of the total sound power radiated by

The uppermost of the bottom pair of curves is the Directivity Index,

Performance in an “average” room

Performance in an “average” room

Now, how do we interpret the

size, these measurements can exhibit large errors.

Simple Spatial Averaging

A lot of Spatial Data

Spatial Averaging in Rooms

Resonances are major problems!

Resonances are differentiated by their “Q”, or quality factor. A high-

resonances are broader when they are seen in frequency responses.

The just audible variations in spectrum

The just audible variations in spectrum

The just audible variations in spectrum

The just audible variations in spectrum

The just audible variations in spectrum

what is and is not audible.

The just audible variations in spectrum

The just audible variations in spectrum

Is there an explanation? It is probably because music and speech are

In order to make any sense at all of a frequency response curve, or a

However, not all measurement systems are equal. Many very

A Totally “Blind” Measurement

For loudspeaker measurements, very long measurement windows are

measurement system should yield accurate data.

AND the most commonly used

A perfect linear system would be described either by a clean

Here we have disrupted the perfect system with a single high-Q

Frequency and Time Domains

Frequency and Time Domains

There is a class of systems that behave as “minimum-phase” systems.

What Systems Exhibit Minimum