By

advertisement
An Electronic Device to Reduce the Dynamic Range of Speech
By
Eric
Submitted
Michael
Hildebrant
in Partial Fulfillment
of the Requirements for the
Degree of Bachelor of Science
at the
Massachusetts Institute of Technology
May
21,
1982
Signature of Author.
Department
I Engineering,
Electrica
.
... ...
e
h.. s.
-
Certified by......
of
Thes is Supervisor,
- ....
Accepted by
-
.....
Chairman,
Departmental
.
Comittee on Theses
I
2.
AN ELECTRONIC DEVICE TO REDUCE THE DYNAMIC RANGE OF SPEECH
By
Eric Michael
Submitted
May
20,
Hildebrant
to the Department of Electrical
1982
Engineering on
in partial fulfillment of the requirements
for the degree of Bachelor of Science.
ABSTRACT
An electronic device to reduce the dynamic range of speech
was designed, constructed, electrically characterized, and
The device's signal
preliminarily evaluated psychoacoustically.
speech into bands
incoming
1) filtering
processing consisted of:
2) infinite
speech;
of
formants
encompassing the first three
clipped signals
the
filtering
3)
band;
each
peak clipping within
filtered/
three
the
summing
4)
and
ranges;
to their original
Intelligibility
output.
the
form
to
signals
clipped/filtered
tests with listeners who had a simulated restriction of dynamic
range showed
large differences in favor of the processed speech
over the unprocessed,
but frequency-equalized
(whitened), speech.
For phonetically-balanced monosyllables, scores averaged 4% and
36%,
and for
words
in sentences
17% and 98%,
for unprocessed
and
However, a simple clipper system
processed speech, respectively.
These preliminary experiments
also showed good performance.
illustrate the problems and potentials of peak clipping as a
means of amplitude-range reduction for the severely
hearing-impaired.
Dr. Patrick M.
THESIS SUPERVISOR:
Visiting Scientist
TITLE:
Zurek
TABLE OF CONTENTS
TITLE PAGE--------------------------------
ABSTRACT----------------------------------
TABLE OF CONTENTS--------------------------
LIST OF FIGURES----------------------------
.4
INTRODUCTION------------------------------
7
PREVIOUS
EXPERIMENTS----------------------
DESIGN RATIONALE FOR THE
THREE-BAND
CLIPPER-------------------
CONCLUSION-------------------------------
-
EXPERIMENTS-------------------------------
APPENDIX-----------------------------------
REFERENCES--------------------------------
io
A
LIST OF DIAGRAMS
33
1.
Single Formant Input to the
2.
Output of
3.
3-Formant Vowel,
4.
Output of
IPC-er with 3-Formant
5.
Output of
OF/IPC-er
6.
Input
7.
Output of 3-Band Clipper-------------------------
8.
Intelligibility Enhancement
9.
Intelligibility of Already Noisy Speech----------
to 3-Band
with a Single
Input
Formant
Input------
3E
to IPC-er------------------
Input------------
with 3-Formant
Input---------
33
Clipper--------------------------
of Noisy
-
IPC-er
IPC-er --------------
Speech------
10.
Audibility
Thresholds---------------------------
11.
Long-Term Spectrum of Babble----------------------
12.
Long-Term Spectrum of Babble-----------------------
13.
Equalized Speech Spectrum-----------------------
14.
Long-Term Spectrum of 3-Band Clipped Babble-----
Al.
3-Band Clipper
P
7
49
Block Diagram-------------------- 5 9
A2.
Clock Schematic---------------------------------
A3.
AC Voltage Threshold
A4.
Pre-Amp-----------------------------------------
A5.
Control knob calibration curves-----------------
A6.
i'st Channel Electrical
A?.
Response of the 1'st
AB.
Response of
A9.
Response
of
Schematic---------------
(
(06
Formant---------------------
the Peak-Clipped
the 1'st
69
Detector-------------------
i'st
Formant,
Formant-------
Peak-Cl ipped,
7o
71
and
Filtered--------72_
Electrical
Schematic---------------71+
A1O.
2'nd Channel
Ali.
Response of the 2'nd Formant---------------------
A12.
Response of the Peak-Clipped,
2'nd Formant-----
A13.
Response of the 2'nd Formant,
Peak Clipped,. and
Filtered----------
A14.
Ai5a.
3'd Channel
7
7(
77
Electrical
Schematic---------------78
Response of the 3'd Formant---------------------
So
A.
A15b.
Response of the 3'd Formant--------------------
A16.
Response of the Peak Clipped 3'd Formant--------
A17.
Response of the 3'd Formant, Peak Clipped,
Filtered----------
and
8-
A18.
Summer Electrical
A19.
Output Response of the 3-Band Clipper----------
A20.
Infinite Peak
A21.
Response of
I.
Schematic---------------------
Clipper-------------------------B.
6s
87
Thomas's Optimal Filter------
-7
INTRODUCTION
range
electronic device that reduces the dynamic
a listener
is to help
a public
address system
other
The
application
sensorineural hearing
conductive type,
to overcome
an
be solved
What
with
is
incoming
a
to people with
an aid
linear amplification
sensorineural
sound detection
discomfort thresholds.
for sound
linear
loss of the
Unlike hearing
attenuation,
range
dynamic
limited
all
where simple
normal
An example of this
area.
a noisy
is for
by elevated
is characterized
essentially
in
impairment.
internal
One
of speech.
a noisy environment hear speech that was
in a quiet environment.
originally produced
is
in
an
applications for
are two types of practical
There
is sufficient
but
thresholds
Thus,
loss
hearing
there
is
a
a problem that cannot
reception,
amplifier.
is a device that reduces the dynamic range of
needed
sounds,
but
also allows
intelligible perception
of
speech.
Two
volume
major
control,
types
or
reduction
compression,
is accomplished
amplifier.
are
automatic
and peak clipping.
by circuitry similar to
AVC part of a radio receiver.
rectified speech signal
preceding AC
amplitude-range
amplitude
Amplitude compression
that of the
of
is used to control
The envelope of the
the gain of
a
Peak clipping
clipping at different signal
of the original
clipping was
used exclusively
abbreviation
"1PC"
stand for
action
the
(as
greatly
the
as possible)
much
amplified signal
hysteresis.
(binary-valued,
signal
IPC'ed signal
the
paper to
:
to
input signal;
times).
with fast rise and fall
i)
and 2)
very
Amplify
Present
little
is rectangular
The
IPC'ed
ideally would change va lue at the zero-crossings of the
preserve
IPC-er
the
are
polarity
of
lag from
input signal.
input to output,
Practical
level
of the Schmitt
which prevents the back ground noise
and
limitations on
imposed by the n oise of the high-gain
This determines the hysteresis
stage,
is
trigger that has
to a Schm itt
input speech signal, have no t ime
an
The
of this
in the remainder
insures that the clipped signal
This
peak
Infinite
signal.
peak clip".
"infinite
of achieving an
One means
the
and none of
in the present study.
be used
will
infinite
In the extreme,
levels.
peak clipping preserves only the polarity,
amplitude variation,
achieved by
are
Different degrees of peak-clipping
reduction.
amplitude-range
is the other primary method of
amplifier.
trigger
from producing
an
output.
Processing speech with an
a peak
factor of
whenever the
I (ratio
of
IPC-er
yields
a binary signal
with
peak voltage to RMS voltage),
input exceeds the hysteresis threshold.
drastically reduces the signal's dynamic range.
limits high-intensity signals instantaneously.
An
This
IPC-er also
The drawback to speech processsed only by
adjustable
level)
linear
is that
amplifier to provide
is
"rumbles"
and "glides",
and the
a comfortable
produce
quality of the sound
section show the historical
schemes
The
intelligible speech.
involving IPC-ing
hearing-impaired.
processing of
(and
an
listening
low-frequency
dominated by
Some type of processing prior to clipping
poor.
IPC-er
intelligibility (Licklider,
it does not have good
The processed speech
1946).
an
is very
is necessary to
studies presented
in the next
development of speech-processing
and their
These findings
application to the
limited
are
also
applicable to the
"clean" signals for presentation
in
a noisy
environment.
following section,
The
brief but
"Previous Experiments",
detailed summaries of the
as background for the present work.
this detailed material,
"Design
Rationale",
taken from the
under study.
presents
important papers that served
The reader may
sample from
or skip to the subsequent section,
for a summary
literature
and
of the main
applied to the
points that were
design of the device
10
PREVIOUS EXPERIMENTS
"Effects of Amplitude Distortion upon
Speech",
This
caused
J.
R.
C.
Licklider
(1946) JASA
concerns the reduction
article
by center clipping,
the Intelligibility of
18
in
intelligibility
and peak
linear rectification,
clipping.
METHOD:
Processing consisted of HP filtering,
followed by
peak clipping.
RESULTS:
Amplitude distortion
detr imental when non-impulsive noise
(peak clipping)
is mixed
more
is
speech at
with the
in
quiet.
a po int ahead of the non-linear circuit than it
is
When the noise consists of sharp high-amplitude
pulses,
peak
clipping eliminates the noise peaks
and
however,
improves
inte lligibility.
Noise added to peak-clipped speech tends to cover up some of
the effects of distortion.
In a quiet environment,
peak-clipped
speech
is better
the
quality
and
intelligibility of
if the speech spectrum is first
tilted by + 6 dB/octave prior to clipping.
"The
J.
Intelligibility of Rectangular Speech Waves",
Journal
of Psychology
Their
speech
D.
Licklider,
R.
C.
Bindra,
and
I.
Pollack,
The American
(1948) 61
the
aim was to measure
and find out why
it
is more
of peak-clipped
intelligibility
intelligibile
in
a noisy
environment.
METHOD:
preceding
The processing
the clipper.
for peak sound-pressure
employed a 250 Hz high-pass filter
Normal
and clipped speech were equated
levels of 85 dB SPL.
added
Noise was
after the clipper.
speech
to noise ratio, with white
following the clipper,
intelligible
vs.
Experiments measuring word articulation
RESULTS:
at
low
noise
introduced
at
a point
showed the clipped speech more
(-10 to +3
dB)
and
signal/noise ratios,
less
intelligible at higher signal/noise ratios, than unprocessed
This crossover was due to
speech.
speech
leveling off
form 50%
intelligibility of clipped
at 50% while that of un-clipped speech rose
to 85% as the signal/noise ratio
increased from +3
to
+25 dB.
The
enhanced
greater power
wave.
low S/N ratios
is
due to the
in a clipped speech wave than in a normal speech
When their peak
greater power
wave.
intelligibility at
in
amplitudes
a rectangular
are equal,
there
speech wave than
is
about
a normal
16 dB
speech
the Effect of Frequency and
"On
Intelligibility of
Tests
in Noise",
Speech
Amplitude Distortion on the
I.
(1952)
Pollack
JASA 24
were conducted to determine whether the superiority
low S/N ratios
speech at
peak-clipped
high S/N ratios
is
a function of the
of
and of unclipped speech at
frequency band of the speech
signal.
then
METHOD:
Processing consisted of sharp frequency
infinite
peak clipping,
addition of
the
noise.
This signal
and
again,
then filtering
limiting,
the
was presented thru earphones to
listeners.
Consistent with previous
RESULTS:
of broad band
considerably
Hz are
intelligibility
findings,
(0 to 6700 Hz) speech subjected
to peak clipping
less than when only the speech frequencies
clipped.
With no peak clipping,
practically no effect upon
is
above 394
the band 0 to 394 Hz has
intelligibility.
In comparing these two test cases the
articulation score for
a given sharp-cutoff filtering condition was compared for clipped
and un-clipped speech.
The
speech
relationship brought out from these
limited to a certain frequency band,
of the unclipped speech,
peak-clipped
speech,
tests
the
relative to that of the
is that for
intelligibility
infinitely
is a function of the S/N ratio
roughly independent of the frequency band employed.
and
is
IQ
"On
W.
the
Power Gained by Clipping Speech
Wathen-Dunn,
D.
Lipke
(1958)
This study measured the
and
5.1,
97 to
portion
amplitude distribution of
When E/Erms (E =
Voltage of
98% of the total speech power
corresponding to 0.1% probability
is
increase
clipping yields a 14.2 dB increase.
to a peak-clipped signal not
peaks.
5.13,
of
These
speech,
levels.
a Speech signal)
So,
which
12 dB.
power
wasting dynamic
is
within that
the peak factor
14.2 dB.
is
clipping (and re-amplification to
peak amplitude) yields a power
due
at various
is contained
(below the 0.1% probability point).
24 dB of peak
the Audio Band",
JASA 30
the effect on speech power of clipping
RESULTS:
in
the original
Infinite peak
increases are
range on the
"Intelligibility
Pollack and
I.
Levels",
at
of Peak-Clipped Speech
Pickett
J.
(1959) JASA 31
Their aim was to determine the relation of
to the
is
High Noise
intelligibility
of peak clipping when the post-clipped
level
held equal
to the unmodified
speech.
METHOD:
Masking noise was
either
speech power
a uniform spectrum noise
between 250 and 6800 Hz, or a low-frequency noise that fell 12 dB
per
Overall
octave between 250 and 6800Hz.
Peak clipping
125 dB SPL.
levels
noise
12,
were 0,
level was 90 or
The
or 24 dB.
over-all speech power after clipping was made equal to that
before
clipping.
The peak-clipped, power-compensated speech was passed
(250
a band filter
through
presented to the
were used.
to 6600 Hz),
listeners.
mixed
with noise,
Harvard PB mono-syllabic
Plots of % articulation vs.
dB of
and
word tests
peak clipping
(at
different S/N ratios) were made.
RESULTS:
With uniform spectrum noise
Intelligibility for equal speech power
1)
independent of the degree of
ratios
it
(-10 to
10 dB)
Intelligibility of
over all
and at
was found that:
levels
is nearly
peak-clipping over a range of S/N
noise
levels of 90 and
125 dB.
clipped speech was slightly superior
S/N ratios and noise
levels
of the tests)
(averaged
by 4.1 %.
2)
For
the high noise
required at
level,
intelligibility
a given
than
level
are
higher S/N ratios
at the moderate noise
level.
3)
Above
a S/N ratio
of
7 dB,
125 dB noise
at the
level,
intelligibility scores higher than 67% could not be obtained.
With the
low-frequency noise,
(articulation vs.
to reveal
failed
the
an
analysis of test variance
dB of peak clippings
peak clipping
as
with variable S/N levels)
interaction between peak clipping
nor was
variable,
a significant
and noise
level found
to be
significant.
"Effects of Spectral Weighting of Speech
Subjects",
Their
I.
B.
Thomas and G.
Pfannebecker
(1974)
JAES 22
aim was to find which frequency bands of speech
provided maximal
intellibigility for hearing-impaired subjects.
The study was motivated by
that normal relative
virtually
B.
in Hearing-Impaired
a finding of Martin,
amplitudes of F1
in
METHOD:
hearing
(1970)
a speech signal
eliminate perception of F2 transitions
sensorineural
et al
in
the
impaired.
Subjects with adventitous
sensorineural hearing
loss were presented with processed speech.
consisted of passing the speech through
The processing
a variable slope HPF.
1600 Hz.
Asymptotic
slopes of
attenuation
octave were obtained by switching
filter stages,
in
which caused spectral
flat,
or 24 dB per
weighting of the 1'st
and a change of slope
levels of 20,
(speech reception threshold)
30,
at
appropriate numbers of RC
(number of
amount of F1
permits the selection of the relative
Sensation
18,
12,
dB cutoff
and 2000 Hz
The attenuation range between 500
formants to occur.
is relatively
a -3
a RC gaussian type with
The HPF used was
HPF stages)
in the
signal.
to the SRT
and 40 dB relative
were used for both modified and
unmodified speech.
RESULTS:
RMS
level
A Gomparison with un-altered speech
revealed the processed PB words 20% more
at the same
intelligible.
sensation
Results were plots of articulation percentage vs.
level measured relative to the SRT.
scores resulted from the use of
levels,
a particular
all sensation
not
the same from subject to subject
subjects) nor even for
the same subject.
subject with the
The
articulation
attenuation slope
although the slope for highest score
for
different
Increased
(different types of
different sensation
"best" slope
averages
at 20%,
loss for
levels
intelligibility enhancement for
however.
is
with
any given
"The
Influence of First
Clipped Speech",
Intelligibility of
The
second
and Second Formants on the
I.
influence
aim was to measure the
Speech was filtered so that
suppressed prior to
of the the first
all
but one
intelligibility
and
formant was
infinite peak clipping.
Second-formant clipped speech has
RESULTS:
JAES 16
intelligibility of clipped speech.
formants on the
METHOD:
(1968)
Thomas
B.
an
average
(over 10 subjects) of 71.1% and first-formant
clipped speech has an
intelligibility
average
of 7.6%.
Spectrograms of the resulting clipped speech reveal that the
behavior of the
isolated formant
There are higher bands
and clipping process.
harmonics
present along with
Tests of the
shows
the original
identifiable as
formant, however.
intelligibility of second-formant
speech were done twice.
subject,
is unaffected by the filtering
a 5.4%
clipped
Comparison of scores, subject by
learning
increase
in
articulation scores.
Most errors on the tests were differentiation problems of
the stop consonants.
Loss
of
intelligibility when wideband speech
be partially accounted for by the
introduction
is clipped can
(due to clipping)
of distortion products of lower-frequency signals.
Also occuring
is the direct suppression of second and higher formants by the
larger-amplitude first-formant components.
"Enhancement of Speech Intelligibility at High Noise
and Clipping",
by Filtering
METHOD:
high
level
Processing
formant,
and R.
Thomas
J.
Niederjohn
16
JAES
(1968)
B.
I.
Levels
Processed and normal speech
of
ambient white noise
at the
involves high-pass filtering,
were presented
listener's ears.
to
attenuate the
process speech before clipping
filter with 24 dB/octave
first
The HPF used to
infinite peak clipping.
followed by
in a
was a four-stage cascaded RC
asymptotic roll-off,
down 3 dB at -1200
Hz.
The SPL of the noise was maintained
without
-5
dB,
for unmodified speech
any added noise was 95%,
Noise was
99%.
intelligibility of the modified speech
Average
RESULTS:
added
at 90 dB.
at a constant 90 dB SPL.
modified speech yielded
an
At
it was
a S/N ratio of
intelligibility score 20%
higher than unmodified.
Under high noise conditions,
intelligibility
is greater than
the processed speech's
that of normal
speech with equal
average power.
With the
first formant suppressed,
filtered speech
are
the
axis crossing of the
largly due to the second formant.
Is
I.
Thomas
B.
J.
and R.
Niederjohn
(1970)
JAES 18
Using their previous paper "Enhancement...",
A im:
start,
in Noise",
Intelligibility of Filtered-Clipped Speech
"The
experiments were performed to determine the
(-3 dB point)
cut-off frequency
high-pass filter
optimal
and asymtotic slope of the
the
that precedes
as a
peak-clipper.
infinite
METHOD:
NO
se
3L xjJ
The band-pass filter for the white noise had
The 20 kHz oscillator was amplitude
from 250 Hz to 6800 Hz.
adjusted to produce
during "no speech"
a 20 kHz signal
Asymptotic slopes of 6,
all
The HPF was
identical RC filter networks
(gaussian
buffered between stages by emitter followers.
filter),
of
at the output of the clipper
(subjective silence).
intervals
constructed by cascading
and the
a pass-band
12,
18,
or 24 dB/octave were considered,
cut-off frequency varied by changing the capacitor values
stages.
material.
Egan's
Both overall
(1948) PB word
level
lists were used as test
of the noise
filtered/clipped speech was 90 dB SPL.
and the
level of the
In comparing the 4 HPF slopes
RESULTS:
and cut-off frequencies from 400 to
dB/octave,
intelligibility score
highest average
12 dB/octave,
slope of
asymptotic
This
is termed the
suppressing
and
dB point of
lower-amplitude,
The optimal
high-frequency components
if the IPC's output signals'
Also,
dominated
by first formant components,
components will
be present
METHOD:
Thomas
in the second
Following
formant
(and
higher) formant
identity of the speech
listener.
and A.
is
4e
IPC-er's
zero crossings are
"Intelligibility Enhancement of Already Noisy
B.
the
then harmonics of these
These harmonics will obscure the
sounds presented to the
in
third formant speech
intelligibility than first
signals.
I.
1100 Hz.
filter has the effect of making the
which have higher
Signals",
an
IPC has the effect of
(HP) before the
output consist mostly of second and some
bands.
and 24
5000 Hz the
was obtained for
a -3
18,
low-intelligibility first-formant components which
would dominant
signals,
12,
"Optimal" filter.
The optimal filter
output.
of 6,
Ravindran
(1974)
a block diagram of
Speech
JAES 22
the system:
has cut-off frequencies of 250
The noise band-pass filter
filter
in the previously described study
and peak clipper were as
by Thomas
The optimal
and attenuation slopes of 24 dB/octave.
and 6800 Hz,
and Niederjohn.
This experiment,
AIM:
before processing,
The RMS level
was 90 dB SPL.
added to the speech
the complement to the previous Thomas and
For unmodified speech,
METHOD:
is
study.
(1968)
Niederjohn
is
which noise
in
noise
of speech
level
(at the summer)
95,
was 90,
and
100 dB
SPL.
(filtered/peak-clipped)
For modified
noise ratio
5,
and
at
was 90 dB.
presented binaurally,
At
RESULTS:
more
few sucesses
heard Harvard PB 50
in phase,
through
S/N ratios tested,
have noted,
replicate
the modified speech was
and
As Lim
to enhance the
as one of the
intelligibility
this simple filtering/clipping scheme has
apparently
an
attempt to
it was undertaken.
Ravindran's
lists
Many more sophisticated systems have
is not understood,
Because this finding
succeeded.
to
headphones.
this result stands out
in the many attempts
already-noisy speech.
failed where
(constant due
Listeners
intelligible than the unmodified speech.
Oppenheim (1970)
of
all
listener's ear
to
at values of 0,
input to the clipper was set
The SPL at the
10 dB.
clipping)
the
signal
speech, the
The
finding is reported
"Experiments".
failure to duplicate Thomas
below in the section
entitled
and
"Discrimination of Filtered/Clipped Speech
Impaired
Subjects",
B.
I.
Thomas,
To see how the Optimal
AIM:
the past papers performs
METHOD:
as an
D.
by Hearing
aid to the hearing
impaired.
"optimal"
filter/clipper previously developed was presented to
The
a group of
so
intelligibility of speech
was compared to that of
processed
in
Filter/IPC scheme developed
Speech processed through the
hearing-impaired subjects.
JASA 49
(1971)
Sparks
W.
speech linearly
amplified
(uniform frequency-gain characteristics).
lists were used as test
Harvard PB word
Sixteen
material.
audiometric configurations were
subjects
with a variety of
tested.
Unmodified speech was presented to each subject
from 10 to
their speech reception thresholds
(re
40 dB
Modified speech was presented at
unmodified speech.
measure the RMS
RESULTS:
In
(SRT)).
the same overall SPL as
The storage oscilloscope
level
at SL's
method was used to
of speech.
13 out of the
17 ears
tested,
higher
at
intelligibility scores were obtained with modified speech
SL's.
in
Except for those obtained
at 40 dB SL,
intelligibility were highly significant.
scores were obtained for unmodified speech
last two cases,
modified speech was higher
unmodified speech higher
at others.
all
these differences
In two
at all
cases,
SL's.
higher
In the
at some SL's and
Thomas and
Sparks give two reasons
why this form of speech
The first
processing should be helpful to the hearing-impaired.
comes from the results of Martin and
Pickett
(1970),
frequency of the second formant of speech depends
If F1
of the first formant.
amplitude
is
ability of hearing-impaired listeners for
is not markedly
an
discrimination
frequency changes
about equal
amplitude
relative
in F2
hearing.
to that of F2,
the
ability of the subjects for frequency changes of
is greatly degraded.
The second
recruitment
(or
reason cited
is
that of the effects
of
the smaller-than-normal
alternatively,
linear-amplifier hearing
speech
is
dynamic
a
The problem that recuitment creates for the user of
range).
aid
absent,
on the
less than that of subjects with normal
added at
is
discrimination
F2
in the
hearing-impaired subjects sensitivity to changes
that for
When F1
which show
at
low signal
increased to
threshold will
aid is either
intensities,
improve
or
an
inability to understand
if the gain of the hearing
intelligibility the discomfort
often be exceeded by extraneous noises.
Thomas and Spark's signal
processing
addressed these
two
problems by greatly suppressing the first formant with the
"optimal filter",
the
and
suppling
a constant SPL to the
listener
via
peak-clipper.
The two
subjects who did not
experience
increased
intelligibility with the modified speech had severe hearing
in the region
of the
second formant and
loss
relied upon first-formant
cues to
understand speech.
"Effects of Whitening
J.
T.
Goodman,
Carhart
R.
intelligibility of
unmodified,
a message
AUDIOLOGY 18
already corrupted with babble.
et al compared the
Young,
METHOD:
across S/N ratios with babble
or
octave multifilter
(30 dB of clippinq)
as noise
introduced before
"Whitening" refers to speech which has been
long-term frequency spectrum could be
shaped such that its
considered flat,
intelligibility of
and whitened/clipped
whitened,
speech processing.
"white".
R.
(G.
This was done using
1925).
Hz were attenuated severely.
The
each reading a separate
talkers,
(1979)
The words were presented to
listeners had passed
However,
"babble" was composed of 5
passage.
listeners
at
an
test at
octave frequencies from 0.125 to 6 kHz.
-8,
O,
6,
Lehiste-Peterson word
and
lists.
average SPL of
a 20 dB HL pure-tone
The
ratios of -12,
a one-third
signals below 250
85 dB.
the
L.L.
To see if peak clipping can enhance the
AIM:
speech
a Competing Message",
in the Presence of
Intelligibility
Young,
and Peak-Clipping on Speech
12 dB were
used.
screening
Signal to noise
Test material
was
RESULTS:
Unmodified speech and
virtually equal
was much
less
intelligibility,
intelligible
whitened speech had
while whitened/clipped speech
(30% vs.
70% at 0 dB signal to
noise).
The
higher
authors assert that Thomas's HPF/IPC
intelligibility scores
clipping system.
probably reduce
sensorineural
than their
Whitening the speech,
the masking effect that
hearing
impaired.
would not yield
whitening/30 dB peak
they claim,
would
F1 has on F2 in the
the background
From
papers described
pertinent to the design of the
dB
(Wathen-Dunn and Lipke,
lower
1958).
intensity than vowels
in
that the RMS measurement of speech
vowel RMS,
almost
amplitude
two facts
important.
are
a peak factor of
consonants,
Second,
(Fletcher,
are about 25
is
of vowels
a range of
and the RMS
of weak consonants.
Studies concerning sensorineural hearing-impaired
they have
threshold
of sensation and the threshold of discomfort.
The peak
voltage, for
a "peak factor" of 1.
have the same output
(within practical
summary that would
mixed
However,
is that
and
if
input
inputs,
predict the
equal to
its RMS
regardless
of the
input
input dynamic ranges).
is not yet
a mathematical
clipped spectrum from the
input
One
some qualitative rules can be applied.
two signals
to
there
is
Infinite-peak-clipped signals
amplitude value,
Except for very simple
rule
IPC-ed signal
amplitude of an
signal's amplitude
listeners
a severely reduced dynamic range between the
show that
spectrum.
which
a measure of
is primarily
peak amplitude
14
Assuming
1953).
we come to the conclusion that there
40 dB between the
and
was constructed
intelligibility than vowels,
much more to
contribute
dB
following observations
is known that speech signals have
it
First,
from other
above,
system that
Concerning normal speech,
tested.
CLIPPER"
knowledge come the
and from common
papers,
"THREE-BAND
FOR THE
DESIGN RATIONALE
a IPC-er,
(they need not be sine waves)
and there
is
a ratio of
'x'
are
dB
amplitudes,
between their
output will
in the
increase
'x'
is,
to
smaller
amplitudes but not the frequencies of the
These observations
leading to
are
the design of
integrated in
is
involved.
signals
following rationale
the
is much
consonant's energy is fully audible.
Thus,
intelligibility the problem
is
of speech that simultaneously does not destroy
In particular,
the
low-frequency components,
Several
and Thomas
studies
(e.g.
infinite
amplitude range
intelligibility.
an
attenuation
intelligibility can be
Licklider,
and Neiderjohn 1968)
clipping of speech can enhance
noise
is preceded by
if the clipping
a
a serious one.
for drastically reducing the
a means
level
as the
Because consonants are so
From Licklider's classic studies we know that
is
the
larger than the dynamic range
vowels begin to become uncomfortable before
important for
is
listener
peaks of vowels and
audibility and discomfort thresholds.
increased,
clipping
we recognize
First,
a severely-hearing-impaired
that the range between the high amplitude
between
'x' varies
signals are
the 3-band clipper.
that the basic problem of
low-amplitude consonants
as
+ 6 dB,
'x'
is that the effects depend on the
Another rule
suppressed.
from
That
dB.
"infinite"
from 0 to
amplitudes
the ratio of those signal's
Bindra
have shown
of
quite good.
and Pollack,
1948,
that filtering and
intelligibility substantially when
is added after processing (especially when processed and
unprocessed speech
are equated
in peak
amplitude).
This
condition can be viewed as simulating the reduced dynamic range
21
28
of the
Thus,
hearing-impaired.
approach
to evaluate
listeners,
comparing
is promise for using this
Thomas and Sparks(1971)
impaired.
with the hearing
attempted
there
such a system with hearing-impaired
the i ntelligibility of processed speech to
Their results favor the processed speech,
unprocessed speech.
but the appropriate contro 1 of whitening the spectrum of the
unprocessed speech was not performed.
highpass filter/ clipper system used by Thomas
and
and Niederjohn (1968).
Thomas
made upon the
improvements could be
It was believed that
and Sparks
(1971)
in the
The basic difference
present approach was to keep separate the frequency regions
characteristic of the
formant peaks,
thought that the spectral
valuable
information for
maintained by
Further,
the
the out-of-band distortion could be filtered out
due to filtering and summing of the
in
same output
input
described
largest
level,
about the
and
in other frequence regions.
expectation was that this arrangement would
as
in peak
three signals.
reasoned that the
independent of the
uninfluenced by components
least
increase
a formant region would be maintained at
level,
literature.
The
allow intelligibility
good as the single high-pass filter/IPC-er
in the
after
The price for this frequency
thought to be only a slight
With such a system it was
at
could be
isolating the formant regions prior to clipping.
specificity was
component
to carry
which are known
identity of speech sounds,
clipping and prior to re-assembly.
factor
It was
first three formants of speech.
system
EXPERIMENTS
Preliminary experiments
were performed to evaluate the
Two
intelligibility of speech processed by various methods.
basic equipment configurations
5
were used
5"Odit
x
E
hJOP&E
These correspond to the situations
is
listener
in-between.
system
in
in a noisy environment,
a noisy work areas
such
aid
as S/N ratio,
a listener
and situation
in a noisy environment
In both cases,
the speaker or
with speech processing
Situation "A" would model
using a hearing
etc.).
in which either
intelligibility
to a
"B" models
'P.
A.'
a listener
(cocktail party,
is measured and parameters
or the sound pressure
level
(SPL) varied to
assess trends and significance of factors concerning the
particular processing method's effect on
intelligibility.
Measurement Techniques
Intelliqibilitu.
(TDH49) were
Acoustics Co.
Listeners
10-2060).
phonetically-balanced
Previously recorded
(PB) monosyllabic words
(IEEE,
Harvard sentences
1969)
and so were
of their
has
listeners scored
informed of their
male
listeners served as
and the supervisor (PZ),
but normal
above 4 kHz,
loss
normal
clinically
subjects:
thresholds
answer
and the nature
on the screen of
(or
its
to obtain
speech,
PZ
frequencies.
A method similar to that
The speech signal
etc).
an entire
The
excluding rare peaks that occur
This value
factor for
at other
a storage oscilloscope for
one-in-a-thousand,
age
author (EH),
(1971) was used to determine the
equivalent for babble,
stored band,
the
hearing.
average RMS voltage level of speech.
value.
own
EH has a slight hearing
age 32.
employed by Thomas and Sparks
the
their
performance
Long-term RMS voltage of speech.
words
speaker of the
Herman;
errors.
Two
28,
The
1948) or
(Egan,
The monosyllables were presented
sentences was David Ackroyd.
with no carrier phrase.
lists of
as test material.
served
The speaker of the PB words was Phil
sheets
(Industrial
a sound-proof chamber
seated in
headphones
wearing binaural
is read out
is stored
list of 50
vertical width of
about
and taken to be the peak-to-peak
is divided by
10
(20
dB,
and 6 dB for one half the
the RMS value of the speech.
14 for
the peak
peak-to-peak
value)
This measurement was
done
31
with
a
(No.1)
list
were
differences
Inter-list
have the same
with a few other
and checked
less than 2 dB,
RMS Voltmeter,
level.
an
or
(white)
HP Spectrum Analyzer
All SPL's
are re.
Our headphones had been previously calibrated,
produce a signal
at
110 dB SPL at the
The headphones were
applied.
taken to
was obtained with
a VRMS/4Hz figure was
(Hewlett-Packard 3582A) when
Sound Pressure
and so were
level.
long-term
The RMS value of the Random Noise
a Ballantine
lists.
desired.
2
0.0002 dynes/cm.
and known to
listener's ear when IV
assumed to be
acceptably
is
linear
over the voltage range employed.
Experiment
I
Measuring the Spectra of Clipped
As stated
above,
the rationale for
Vowels
the 3-band clipper comes
largely from the desire to process the formant regions separately
so that formant positions are maintained
range,
and
their
this series of
spectral
peaks
automatically equalized
measurements
measurements of
steady-state vowel
in their frequency
is
in amplitude.
illustrated this approach through
differently-processed vowel
sounds.
was electronically synthesized using a Bell
Telephone "Speech Synthesis" kit.
This device consists of a
square-wave buzz source followed by three cascaded formant
resonators.
In
A
in
by the clipper.
stated
above
about the
The effect of
is shown
seen that the first
formant
in
Section 2).
"Optimal
The effect of this filter
and second formants more
bring the
These examples should
is to preserve formant
locations,
illustrate
it
a 2-pole,
Filter"
is
to
a result,
in the
and 2'nd formants
clipped spectrum in Diagram 5 both the I'st
maintained.
third
amplitude of the
As
into balance.
is
IPC'ing on
it has passed through
so as to
it
the second and
is preserved but
(Thomas and Niederjohn's
attenuate the first formant
first
a
Here
and 4.
in Diagrams 3
the three-formant spectrum after
described
IPC-er on
an
Diagram 5 shows the effect of
are obscured.
1100-Hz HP filter
the smaller
larger components suppressing
three-formant vowel
formants
expected from the rule-of-thumb
is
This effect
in a clipper.
components
the resonance has been sharpened
Note that
Diagrams I and 2.
is shown
on a single-formant vowel
a clipper
The effect of
that,
are
if the goal
is desirable first to
clip each band, and
separate the formant regions by filtering,
then filter out the out-of-band components.
The
and 7.
action of the three-band clipper
The formant peaks are
(approximately),
and equalized
maintained
in
is seen
in
Diagrams 6
in frequency
amplitude.
Note that
first formant region the maximum component has shifted
next
lower harmonic.
in the
to the
10/.
lX 10OTO THE CENTIMETER
F1K"#E KEUFFEL & ESSER CO. MADE IN US
18 X25 CM.
46 1523
A.
., q N-r
-
-
L4 }
-:
-I F4 F
E1z
1
_
1 L
LL
H-
dfl
ji
H'I-
I
FF
IH -I-Iti
k
I''1
1-
H
4-r
iFIIr
4
T-F;2
II
'
+-4 4
4t
14
I
##41F
FL
LI~-7
-47~
+4
Li
[4I
4
~
-I----- F[
A41
T
IfT
I
0
I 1<
L!
I
2f36
:3
10~.' KEULFFEL
lX 10 TO THE CENTIMETER
& ESSER CO.
IKF---E
18 X 25 CM.
46 1523
MADE IN U.S.A.
I W ff5 V/ OI,&~Z
o~A7p.4r
t+t
Lr
~
1
rPc.
r6
4iIf
4.4
"1:
11
t-1it
_L~f
I 14
Ij
-1
-II41-
-
14
#4j4
T
-
ITj
4ijjj
4
rj lj
'J
+i
7+;j+
4T
_E*T111
4-Tj:0
T
T-
#1riI 51C7?
0X
-
L
46 1523
18 X 25 CM.
10 TO THE CENTI METER
lO#E
II~~KEUFFEL
& ESSER CO. MADE IN U.S.A.
r
F1_4
jj
' 7
-H
I
Lt- - +- -1-1
L
-4- li_
-it"Li----
I-
I
JLL
:_L:_
i 1-t I
-I I I I
J-1,
A44
t
-i I
L-L
F! J-
- _Ir - -
--
- -
r-T-
TT
II T
ELL
-7-
L
1
I IN[ I I 1 1
-
T-11 Ii; !I
F11-
I
T
Fr 'r
Tll
14-4-
L
111
-4+-T-
1
L
7,
IT
-
t
tt
-Tj 44
Tt
-1
I -
r::::
t
F
41
II
V!
a!
#
j-
44
-14V,
-:444 t- --44
I
_r
r -It
T'r
4
T[T
WIT
IT~4-
7
-tt
+
-1
Tl
II
T
J
14
-i-!-
I Ml
iT
.....
I
I +
j-1-
r
.
-+
_-r
111, ITTlittij
LE
IL El
A
-TT
'4
_T .
T
-ITf
+
LF
4T_
-
L
4-
+
44
4-H
1
71
4,
4.
r
t
-4- -4-4+
_7
Ji
H
t
1
-1 ,
i
-44+ =rT
7__
F -Fi
_HT.-4- -1
41
1
I- I1-
'i: L j I-T
R I TR
1
111
EalT-L
"Tri
4 H+ H
t
_IL
41I
--
T =r-
4-
+
L
L
+rt-r
L It,
-I
J_
IL
tE
ITq
+L
4"
t
+
+Lj+
44
_I ir_
L
I L
J, "1T
[11 4
4-1p-
-
4-4
_T-
L
L
i
44
-IT
T
V
f
-
L
L
IN
-
t
-
P Tr
FF IF'
ff
I,-
T
IT
T
TF
-
-1-
I
-
+
LrE
q1111
A _L!j I Fl_
-
-- -
4+ 4 4 It -H
t ij
L
i-H+1 I I
+
I
- -- -
_J
#
It
-1-4
-1
L
L
7.
4
T
41
+
I
t
T
I TIt
I
L
L! I
-
I-L
-
44i
4
T
L
T-I
77
L
ff
t-I
1
-4-
A_ -
TEF
FITT ;
14
1
Jt
I+
_E_
T 'tr,
.....
..
j,
q
L.-
;dpft "L
d
A
m%1
I.,
1\
IT
L
b jul -*
oop
LT_
10'2
18 X 25 CM.
lX 10 TO THE CENTIMETER
& ESSER CO. MADE IN U.S.A.
im~KEUFFEL
46 1523
74
-r-
T
TI
rt
T- f
j
L
1-1-1F- -1-1
,
T-
j
-4
L-- +1
L
IT
jI
F
-
I
--
L j
IF
-1
-T -i-
r
ILL 4
1
4#
T Tl:'-I- +T+TF 1: T1 --- i
-T
IIIT
17T
14
4+
41-
F-l-l-
i
ITT
-1 I-Tf
r
-E
--r+F
r
T l-
r
7
------ -
it
H+,-
I
--,tT7
LL
Ad-
:L
L
1
1
L 11
f#
-L f
kL
-1-
f -H-
I-
-41
-i
1-14-1, 1
ti-
T-
1
14 F
+w
A
-1
-4
1:
-T
-- F
If
f7
J11
IF
-
i
T
flF
L
Tit
t
A
---
tp
t
t-f-
L lL
r
-4-1
4-
. ........
-r
-T
it,L
iL
+144-1
I
0
F
A
_4
-
+-
+
7-1
L +
TF - -
"Er
040
I
-
jji
7
4+-
-rT1 i
6
-+F
Th-
i r
I I itLL
I LI T
It
-
1 FIF! I it 1 F
1
4#1
-T
i
41
ljj-
-
-r- -41-r- -r
Tj
1-H
TT
Ft
1
-f-rt-r -t
: IA:L
,411- -
T
-r
-
L
V11,
P
Op"I V
T LL
tt
II
I i
t
LLLJ
-4-
TF
Tr
-
4 1-4
IL
-4
T
F1
+
-
L4 L I V
FF
1 74
-1
jrL
Lr r
i+4
4
II
F'T
IF
i 1 1 -4-44
L
4-4
4
4TT
j-
11
r
L4,
+=
L I
FFF
......
.. . .
..
77
rf-t-7
i-
-W4
4
-7
T
L
,
A E -it- i
4
,
1 T
LLL
I
I
1
1
T
Aj
I
I
IF
+
J
-H
r 14,
I
-
t t t t
I,
-l- _4
4
,
tj
F4, 4-,-t -H r i
Ir
+
4
4
:-7-
L/
0
I k1
Y-
- .5
s
d
Ica-2
414CA
18 X 25 CM
46 1523
err,
ovrp(4T V+ C'ti#.ep witA
L
-1LI
1-1
TFD
T--T-
4
i- I
I- -i
1
r
-1
-W,, -4-4- 1,
j4-
-F
+i
di
-vt
-
-- lT--FT
I i t;
- t-t4 f
t-
a
Ta
+4
I
4-
Fr
-t
'Lj
-i-I
IT
T_
I+
i
T -
-1
- I
- IT
I
I
T
T
IT F
-1uFT
-rE
4 _4
E,
I :I
I
T-1
-4 T-1I
- _:
Fri
i4 ;1
T
7
FF
-
L +LHFL
-
it!
-4
q
L
+ T
-f-+4
4-
F r
t
i1
1 14
4 -
-
i
A I I L
111
A,
H''
-f
V i
1
L
-
A
1
I
fl
Lr' I-114-1[
- -,
_T_ T
t 1T
hu
14*
_rf F
-
41
[111,14 '44
71 -
-H+
Li
F
14",
L +
t+
#
10OTO THE CENTIMETER
KEUFFEL & ESSER CO. MADE IN U.S.A.
T
t
L t
FE
1
L Ll L, -fL
I r
7-
_q
+
44-
7,---
HE
I
_T_
r
77-F
-777
4
4-H
i
-
T
Ii I
: i I I I Ir
it
L
-I-
_LL -
ir
I
t
[LL -tl,: LH
ir
_TT
It
ilT :IV
I
I
d
F t14 t 4 it
IL:- L
L
IER
I
ff
--
7+
1 -tAT U F, t T
-1
11- I-PH
I-
i_[
if
. I
L n-)
-1 T
UL
:
7
+
In
i LL
-H_H_
4-44
+
LT_Ll
it
I ii i I
-7 7'.
-i
T
i
]4
'
I f 1~+
,
_L7
i
FITit+
I
_F 1
- - I
J
AIILLI
4 [+
A- _+f_
L -4
T_
IT-
IL iH,
LAU
i1j,
I- -Ij
A-1,
-A
H-
ftF
-
4
-___LL
t-
i 4
I
FF,
it
+i
H
F
4
IA
_. ': -_
ELL _Ft_
i -
i1IT
L
-1 1
F
[I T i
4
L
IL
L
T
1 f
r
41JV
11,
J-T
r
-1-T IT
1:
T t4Tt
t-14 1-1-1 1-!
J
1-1-
4
t
+11
Ll
L _LLL
L*
Ir-4-
L!
iiiiiiu-
j:li I
t
717 7Hi:7
- . _. __
_
_u
j,
-71
-
V
1_L -_L
'I LT:
L.
FF
I !--i-
+
L
+
L
h
i
Tr
L
v
it
L
T1
L
_7j_
-
L
1K -,El410X
A It
. :1
F
'r
I
1+1-
-L 7
-17
ii
-4
L-4-
-
T
-1
4'
41-4
E
-
"I LL j
T-,
rT-
7-L-
-t-
--
-T-
17;
!4T[
-i
+
!+
L
j
;:d:
4-4
#
1,-
-JLL
+--4
4,i
7
--
L
77T
LL
44
p
Ill
1t
4L, -I------LL
-4
it 4
p 4 14
#
#
;t
-:
L
+
4
Aq
t
L
4-
T
...
r
4- 11
-H
11
IL
---
L ltt
+1
4- t-l -171-t
-4
L
-1---
L
Ll
Ek,
J 4-1
r
rt-
.4
:
4
+
+
th
IL
T7-
:tit
- i ---
-
4
t
4d
FL
+T7-
AL
id-I
-T-
TF
H
-j=r111 - l
44T,
EE
#4
Fj44
-H-
1,
L
H
+
H
I T-
I
T-
I-A
t-
1
'S
+
-T-
t4 tv T-
-A
11
-- L -
14-
L
iILI
T
7
L
-L
'TT
r- -rr
L
4
J,
::
:
T-A
LH
-T
-7-
::
-E
-ZfE
L-1
-A
L
i M-
Il-
:
_-L
7
JJ
-
4
Ll
6-t-F-
7
L
11
L
11
T
-44 1
-1 T4
IL
1-4-
I
-
J-1 j
-i i+
+
+
7+l"
. i
L
+4
T
T -Ir
+
r
r
-LL
4
77
IT
Prl
tt
-
-4-
tt
4-
4
4
77
-- I- --
j-7
LLI
t
-- t-M-1- 77-
-- LLLL-J LL:-
-
+-
-Tt
=1TF
HET
L-d
L
-7
7:
:
F r
Tt
-
L
T-
ILL
-T-
-T
-- ,+
4
1 1 L --
I i-E
4-
LLL
IL
LEELEE
:T]-::
1
4
I
E:
t
.
-
H-f-H-H-H
L
-J-7
T-
I- I
I-
I -
i
Poll
-7
H
4--
7
-4-
103 OX 10 TO THE CENTIMETER
& ESSER CO. MADE IN US.A.
18 X 25 CM.
IFEiFFEL
TI
_
r I
AI-IV
L
t~
+
d
t
1i
H-TttfI
L~ r
-__
-17~
W
TI
L
ti
T
I
I
4
Ll
I rI
I-
-
1144
-rt'
~ IH1 ihI-i7-
F
['1
-L{-T EL
FIIT
IT
-HK
17i
I<
TH-
-F,
-4
:4
IIIpi
IT
I
r!1
1 11,
-
~-
A_
14V~FFfi 2 1i F~ V T~.
t.
L
I IVFVV
-1 -
!t
- -I
I
F
L
I
T
I
I
I
7
I T~4
T
L;VA
H
L V
-fi LfI 1
<I
I+
TTTT
1
-
-
If
-
L H4!V2 14 1-i
46 1523
~
+
FI
17l
Vi V
itF 1
I L~j
ITI'4T
2
<i
II-IJ4iI
I
K
I
I
i-I
q c
Experiment II
Replication of Thomas and Ravindran's Experiment
METHOD:
(1974),
The experimental
procedure of Thomas and Ravindran
and described in the Previous Experiments section,
was
followed as closely as possible.
RESULTS:
Diagram 8 shows the
intelligibility
(%)
vs.
S/N
ratio (dB) for curves representing the results of Thomas and
Ravindran,
and the present study.
The results are very different
between the different experimenters.
Thomas and Ravindran showed
the processed speech more intelligible than unprocessed, and we
have showed the un-processed speech more
processed,
intelligible than
at all S/N ratios investigated.
speech RMS voltage was
If our measurement of
inaccurate, then our data could be shifted
horizontally to compensate.
However, the fundamental differences
between Thomas and Ravindran's results and our own would still
not be resolved.
Experiment III
Intelligibility of 3-Band-Clipped Speech that
Corrupted with Babble.
Block Diagram
is Already
(
pdj e q
Pep~icaA-io>
o
Z-.Th4*sf
'1>rlel/lrj
- 2>
(9tell)
AES22((7
I
7
/o(-Ty -Teo'
-9
,~4/s~
67?
(,4eee{, Acro-04
7
d+
/
U
~t1
A
P.7.
7R~
I)
,ioifiec/
f ud
/
240
Fil4e
ed;
/ f&WdM
Filkrted, Cippxc( Efi.
30
/ 0
S/N
-a4lc I-;kl WR
0~ ch
0
171
/0
/5-
The formant-amplitude-control knobs were
METHOD:
informally by PZ for best
intelligibility
First Formant 56 (-16 dB),
Both
the
LED came
indicator"
"peak
input
the signal
Second Formant 50
input gain knob
The
(-22 dB).
Formant 65
on very
was
other
are
The settings
(-16 dB),
Third
set so that the
infrequently.
RMS voltages were
and babble's
to the summer by the storage-oscilloscope
at
measured
method to
The HP attenuator was
establish a zero dB S/N ratio.
The
and quality.
settings acceptable.
(EH) found those
listener
adjusted
then used
to change the S/N ratio.
Phonetically-balanced (PB 50) word
Long-term SPL of
1948).
80 dB for
all
RESULTS:
point
('X,
list) for
or
the speech signal
ratio
Diagram 9 gives the
(Egan,
at the earphones
intelligibility results.
'0') represents the score
either EH or PZ.
is necessary
of
was
An
(one
is always
less
increase of
intelligibility
Each
a 50-word test
The modified speech
for equal
unmodified speech over the
Both curves
were used
tests of this experiment.
intelligible than unmodified.
S/N
lists
about
of
10 dB of the
modified
and
intelligibility range of 20 to 65%.
in the diagram have
roughly the same shape.
P
4/b-
vI
s
A/es
-Jyox+
raU
S
-je
90,4
1)nc r'/
96 4
e
4
4
k
p
ic
S/Al
C
-L
0
(d)lt,
/0
10
20
Experiment
Processing Schemes for Reception
Comparison of
with Limited
was
by Listeners
Dynamic Ranges.
Block Diagram
It
IV
ase-
desired to test the various speech-processing schemes
listeners who
have small
However,
dynamic ranges.
with
for this
preliminary testing we chose not to test hearing-impaired
subjects.
Further,
a masking
noise could not be
the detection thresholds of normal-hearing
the dynamic range to 20 dB because such
painfully
intense.
Thus,
exceeded
METHOD:
was
a specified
pure-tone
thresholds
simulated
light when the
level.
in the normal-hearing
a random white noise mask.
Diagram
(generated with the MAICO audiometer)
in headphone voltage
analyzer) vs.
and "noise masked" with
listeners
10 gives
detection
(measured with the HP spectrum
frequency for both PZ and EH .
thresholds were performed with
only),
a noise would be
indicator that would
Threshold elevation
produced with
subjects and reduce
the discomfort threshold was
artificially with a visual
signal
used to elevate
These detection
"quiet" presentation
(pure tone
a noise spectrum level
of
p (k5 le Z/ ;-
q4'
Yask W wlA4
- 3a
A 1/
I
;eWI3 ~T'~
>vle/5
i
70
4106
4
3e-er of
A
A4wll
/
-
TFAI-esYs
v!.;
01
-Nl
-Zc-
p
250
5o
1k
Ik
4/4ye
8k
8,k-
+4e e4ts
q6
fzp
-80dBV
our detection thresholds
elevated
To
that would
imposed artificially.
-35dBV
(as
diagram.
a "discomfort
This
light
performed,
For the tests
exceeded a
either -45dBV or
when a 1 kHz pure-tone was at
This
measured on the HP spectrum analyzer).
represents useable dynamic ranges
of
about
indicator
"pain"
this
an LED
was
threshold
light whenever the wideband speech signal
certain voltage.
to
in the
as shown
simulate a reduced dynamic range,
threshold" was
was set
This masking noise
which converts to 30 dB SPL.
15 and 25 dB,
respectively.
With this
unfair to
artifical
compare the
intelligibility of speech processed by
low-frequency
long-term speech spectrum only the
portion of the spectrum would fit
into the
the
Because of the
3-band clipper to that of unmodified speech.
slope of the
it would be
range,
reduction of dynamic
An
dynamic range.
appropriate comparison would be to speech that has been
spectrally tailored to fit
in the
Since detection thresholds are
threshold
is constant,
whitening
of the
listener's dynamic range.
and the
nearly constant,
the necessary tailoring
long-term speech
spectrum.
amounts
The
"pain"
to a
frequency-gain
characteristic that whitens speech was determined by measuring
the spectrum of continuous babble
and
adjusting the
G.R.
1925 one-third-octave multifilter
13).
The
speech for
same settings of the
the
14,
(Diagrams lip
12,
and
multifilter were used to whiten
intelligibility tests.
be compared with Diagram
settings of a
the
This "whitened" babble can
long-term spectrum of
KEUFFEL
&ESSER
7i~
ff1
TFI
Lv-+ttt <1T~
IA I 1TTL1-i
if
V1 h14
MADE
46 1523
INU.S.A.
jf
FF1
411L~
j
CO.
fjIF1ft
1i T
1
7_
«Ti
'--F
III1
ILL
-
t
-
-f
1-
7
Lit
-Tf -1- 4~
-
Hi
41 17<1
I7 1
I
I
-
-I-~
117
1-
-f1 - 1-7 - 1- -1
T-iT
-H
L;I
7
f
<
7
1A flf--
#11
{I
-f7 Ill~f Hill
-4i~ 1Ft~
F1
i Ii~
i F IF l
iK
1K
Il
1
t
<
7
I
1y~
t FFt4
ti 71ff
r4I$~~~~iiif~~~~t~~
t
F
I~4
_41,7 1 1
Ti ;+ P-1
AIF I1
JIt
k
1471
i
t~11
---F+
~+4
I4
i-L~
I
-l
t
,
LI
~
t-1
<t
FFT
+
tiVI U
F<E
lA
4A 1
#-
ILILTtt
-
-a~I
FF~
+iIFt
F44I
r~71
I F
4_
-t4-
~I IFI
7,~~~~H
l+
F~f7TF
L111~
f
I
df;IFl A:
i
lt-
FLL
~
74 14
4
4A4
7
1
411
1 1
Sawly OL/YL+v
/
46 1523
Mealik&10 TO THE CENTIMETER
1Ks10OX
KEUFFEL & ESSER CO.
MADE
18 X 25 CM,
IN US.A.
7-5-4
3oI
5a-vftpe 5
(fVtt
0 aSziA
4BaV
41
It_
1
i#:
!- t
4F111
L~-
:: 12
i
~I
-1*14t1-
-AE
L-Lf
1 1 1 i,-L
1
~-TF
t-H
-16 49V
I
I~I
VhT
4~~.il
fTt
F
4
~
4-i
I~lIftV-!
I
1J Kr
FFE->I<
__A
IIL
4
II
Hf-
F
~ I ~F
H
1- 1
u~ pT1b~~
I
.L
-LH
2
itjliLi
IT
Ll
1
1' i-V ,q
'
J- 1-
-
TlttI~
1
-
il
Ir
~j
jL
_4
-177V
fI--
tf
10'
IK,-"-E
lX 10OTO THE CENTIMETER
KEUFFEL & ESSER CO.
~IJ4OL J1a~m
I
~ FFIF
I<
41
W-1
~
-i~'
111
IA~
J
1-
4Fk-
+1~ 7'T
~ ~ ~~~~
FF
Ii
I-
+1
tf
V II I I
LI-
1
1
-KILT
1F
I
1
+ F -1
V
j I II~
l+
Fw1tiiVVI~IHI
I
L
4
~-2 ~4
~4tLL4~24~44
IIIFFITI1
F
F
-1 7I I -1
Il~~j I jTh
--
H
~
I I I I
.. 1+----IT
LL IH fHi-- t I ILL- 1_tPL
i~I~t~I-U
- +i
F111FIF_4 Hf
7>> 'i
-l
jZFIFA~iVL~L7
TLF- 4+
I I I II -
---- I
I II I
I2>4
*+
w
J l#I~I
t
r~~~rn~-r
144~t~
1
Ii
~X ~H4V'44' ,
1,41
~+F
I
~1:61; il 1112011
~
~~~F~~1-F1
V,
1
VF
F~~~7I4
t+
I
~iI~
T K1
I
-- 14j
>
IFF
I
-I1F
I
-fV-4i
1
-f-f
-,
I
I-HI I ri-F
IT
L
1.iI
-
+
LI[
46 52
44
LFHHLiF
+
T
SpaX-414:
'i 4 L
1I:
~
18 X25 CM.
MADE IN USA.
F
I
~
'tj
F~ __
L
4T-
-0
I-I I-I I
H
~
:1,1 14
+A
F
1
IFF
1:4-7
F--fl
4r
[I
T}ITIt
-4 4
~l
f
j
1
]TF
4 1
F
>11
TI
T
TITW
I
IT
++1
18 X25 CM.
46 1523
ec
3-v4 4a
CIWepr
Eot-,a~PL4cotS-ot5
- i-4~ 1
'I
F
II
H
i
4 -I4 F
I
-
F
F1
1
i
F HI
~l
I
I
IT IL
I
t11
- _T
1
i
T 1' ...
_4+
14H-&
F
iIFL
LL
1-
t14 U41LtL
F~lLIL
Vi
F
-14
iwI
l
I
{L~~~~~
I
to
(
10
X 10TO THE CENTIMETER
IF~KEUFFEL & ESSER CO. MADE IN US.A
1
LLJff4
41-F
-FI
--I
*l
III
141
F4
i-
L
714
1'K > A
F-
"L
H---7
-
_-
tV
7-1f7~
+~
I-1
,
I
-'
F '
ri-
tV7 1L
I4
FFL
_II' -,k'Fi I--
- I-
,'_li-i--i
-IFI
F-I,
f74!
-1~
513
3-band-clipped babble.
The
third system that was compared was the
(OF/IPC)
Infinite Peak Clipper"
The masking
and
described by Thomas
and described here
Niederjohn(1968),
"Optimal Filter/
noise RMS voltage
in Section
2.
(which caused the detection
threshold shift) remained constant throughout the experiments.
When different devices were chosen for speech processing
(one-third-octave multifilter,
different si gnal output
attenuato r
the LED pain
located
),
voltages would be pres ented to the
To compensate for this output voltage
summer.
an
3-band clipper, or OF/IPC
in the sound chamber
level difference,
was adjusted so that
indicator came on only very infre quently.
This
adjustment w as done using continuous discourse speech material,
listeners,
that
to be altered during the testi ng sessions by the
allo wed
was not
the pai n
meant to
has no effect on
and
the pre-set vo ltage threshold
i ndicator is sensitive to.
simulate a hearing
aid user's
This
adjustment was
ad justment of output
level
to prevent pain.
Intelligibility test material
(Egan,
was either PB 50 word
or Harvard sentences (IEEE,1969).
1948),
were EH and PZ.
Each entry
in
either 50 PB monosylabic words,
The
lists
listeners
Table I presents results with
or
10 Harvard sentences.
The 15
dB range was
not employed with the PB words because nothing was
intelligible
after
triggerings of
adjusting the overall
the "pain
indicator".
level for
infrequent
This was true with either
whitened speech or
not
3-band clipped speech.
investigated with
The OF/IPC system was
PB words.
Table
I
PB words
Dynamic Range
3 band-clipped
Whitened
25 dB
PZ
EH
PZ
EH
4%
4%
33%
40%
Harvard Sentences
Dynamic Range
EH
PZ
EH
16%
98%
98%
96%
96%
0%
6%
2%
68%
54%
EH
25 dB
18%
15 dB
0%
results clearly
indicate that speech processed by either
the OF/IPC or the 3-band clipper
whitened speech.
OF/IPC
PZ
PZ
The
3-band Clipped
Whitened
is much
more
intelligible than
When the usable dynamic range was
15 dB,
OF/IPC
was superior to 3-band clipping.
This
OF/IPC having
and 3-band clipper signals
a peak factor of
1,
is
apparently due
to the
having
a peak factor of about 4 dB (measured on the storage
oscilloscope).
clipping
The
larger peak factor
in the 3 bands.
added together,
which
also
is due
to filtering
These three filtered signals
increases
the peak factor.
are
after
then
L
CONCLUSION
There
achieve
are many types of processing schemes that could
"figure of merit"
keep in mind that the relevant
system
It
amplitude range reduction of speech.
is the
the reduction
important to
for such
amplitude range.
processing
This study was concerned with two types of signal
that were shown
speech
to effect a reduction
without severe
loss of
in the
amplitude range
intelligibility.
(1968),
of
signal
The two
processing systems were the OF/IPC system of Thomas
Neiderjohn
a
the processed speech relative to
intelligibility of
in
is
and
and the 3-band clipper developed here.
Speech
intelligibility with these two processing schemes was compared to
that of either unmodified,
or frequency-equalized (whitened)
speech.
In the first phase of the study
intelligibility of
were compared.
Thomas
and Ravindran
listeners
at the
(1974),
which showed enhancement
already noisy speech.
in prior studies
Ravindran
inputs to the various
1974) were
of
With this system,
(Licklider and Pollack 1948,
able to
Thomas
achieve close to 100%
intelligibility of processed speech with high
ratios.
systems
This study was unable to reproduce the results of
intelligibility of
and
noise
the effects on
(>25 dB) S/N
However, we (PZ, EH) were able to achieve only about 75%
intelligibility with high S/N ratios.
So,
perhaps the
reason why
we were unable to reproduce Thomas
and Ravindran's
that we
To
resolve the
inconsistencies brought
out here,
of that system should be performed to see
better with
recognize the HP/IPC'ed speech
further testing
if people
listening experience.
literature with experiments that had noise added
In this case,
processed speech.
the
intelligibility
processing schemes effective here were the
3-band clipper.
allowed
the range
was
With
a listening dynamic
level of
about the same
of processed
The
lowered to
range of 25 dB,
the two
When
the OF/IPC'er yielded greater
15 dB,
larger peak-factor of
OF/IPC'er and the
intelligibility.
intelligibility than the 3-band clipper.
to the
in
to the
vastly greater than that of unprocessed speech.
speech was
systems
learn to
can
able to produce results similar to those reported
We were
the
results
IPC'ed speech.
are not yet well enough trained with
is
(1974)
This
is apparently due
speech processed by the 3-band
clipper compared with the 0 dB peak factor of speech processed by
the OF/IPC.
In the Thomas
system was
(1971)
and Sparks
evaluation,
found to be of great value to several
hearing-impaired
listeners.
Two
low-frequency hearing produced
listeners
their
HPF/IPC'er
of their
with only
lower scores with the clipper
system than linear amplification (uniform gain across frequency).
Perhaps the
has
3-band clipper could be of assistance here,
i'st formant speech waveforms present
in the output.
since
it
Because of the positive results shown
by both the 3-band
clipper and the simpler OF/IPC-er with simulated reduction of
dynamic range,
future experiments should
intelligibility with
hearing
loss
listeners who have
include evaluation
severe sensorineural
and very narrow dynamic ranges.
of
APPENDIX:
input
devices.
The pre-amp's
output
into the
signal
band =
A
in diagram Ai.
the output
indicate when the
accomodate
of the
pre-amp
pre-amp
is
is clipping
introducing unwanted distortion.
thus
input
Monitoring
indicator" to
an LED "peak
and
shown
pre-amp provides adjustable gain to
input
different
is
block diagram
The functional
linear
57
ELECTRICAL DETAILS OF THE 3-BAND CLIPPER.
200 to
is
applied to filters that separate the
(i'st
first three formant regions of speech
= 900 to 2800 Hz;
2'nd band
900 Hz;
to 6 kHz).
Audio taper controls allow adjustment of
each of the
three bands.
=
3'd
the
2.8 kHz
input to
Ten-turn potentiomenters control the
cutoff frequencies of the filters.
outputs of the
The
peak-clipper,
by
consisting of
a Schmitt-trigger
filters
initial
an AC amplifier
(Hysteresis
off
Each
the
(gain =
'dead zone'
insures that the clipper's output
squares
are sent to an
=
12),
74 mV),
is either high or
infinite
followed
which
low (i.e.
signal).
IPC-er's output
is filtered to eliminate out-of-band
distortion products.
Finally,
a summing device
filtered signals together.
ohms.
adds the three
Output
filtered/IPC-ed/
impedance of the summer
is 30
75
im.A
L
14
p
p
2-d) Forn7t
7
L-O---*B
I-L7
P F/
F9
D aj-
-P
I
rOu,
f
5
)Lt
2S<9o34
0
k
452-0 1k4
of the output signals present
All
points
i thru 9 on the block
labeled
diagrams)
are short-circuit proof,
at Banana jacks
and electrical
(test
schematic
in the sense that no damage
will be done to the circuit.
Performance Specifications
Positive Supply
Negative
Typ.
8
9
-8
Supply
4
Clock Supply
Max.
11
-11
-9
Volts
5.3 Volts
5
Negative Supply
100 mA
Clock
100 mA
10k ohms
Impedance
30
ohms
Output
Impedance
Output
Impedance at test points
5k ohms
Output
Current
18 mA typical
Output
Signal
I VRMS
3mV
Output No ise
Input Dynamic
Volts
100 mA
Current--Postive Supply
Input
Min.
range
(with
gain control
Max.
's
57 dB
at a fixed setting)
Band Frequency Ranges
i'st
200
--
900 Hz
2'nd
900
--
2800 Hz
3'd
2800
--
6000 Hz
-4 dB
Output Signal Peak Factor
Component Selection Criteria
Because sharp definition
bands
is
Mos-monolithic switched-capacitor
and G Reticon"
sheet enclosed pg.
R5609 (seven-pole,
pole Chebyshev)
Thus,
of formant
) were used
low-pass),
"LG
(data
filters
The
as formant-band filters.
six-zero elliptic
were used
important,
and R5611
(five
in series to form a band-pass filter.
the cutoff frequencies of the pass-band can be
independently
The
Reticon switched--capacitor filters have the draw-back
clock residue
signal),
residue
adjusted.
(25 to 100 mV rectangular pulses
and DC offset ('100 mV).
is removed by
In this
a single-pole RC
DC offset by AC coupling.
riding on the
application,
low-pass filter,
clock
and the
of
The other
semiconductor)
Amplifier"
critical
--
active
is
component
a "Wide Band,
the HA -2605
Operational
High Impedance
(bipolar monolithic).
It
is
(Harris
internally compensated,
and the following specifications show its
suitability
in this
application:
Impedance 500 Mohms
i) Input
2)
Gain
3) Slew
(DC)
150,000
Rate 7V/uS
(to 0.1%,
4) Settling time
5)
Output short circuit protected
6)
Power:
These
choice
as
imput
Its
15V.
a low-power high-quality
impedance and gain
of feedback component values
(cascading)
and
ease of
allow flexible
interfacing
for multistage design.
Analysis of
I.
with supplies at
features allow its use
amplifier.
audio
90 mW total,
1.5uS)
large signal:
Individual
Circuit Blocks
Clocks
Diagram A2 shows one of four
drive the Reticon filters.
output with
fast rise/fall
identical clocks necessary to
The clock should have a sharp-edged
times
to provide precise
definition
of
/Orc
O Ev/o
r- C/c~i~63c,/Q-/~S
A/7/ I Z.5,VA
out put
I
5
4
CD
o
Teg r->it
A2.
a5
+IS.
~45
i) C,~[e4',44)722Aer~
0I-e- /o k
A
3)
Ck-$
27oo pR
"I
/0 1'urrE
N/o 1"r?1 -),
Ce rne
4|tNrej.e cf
ReLbkMe>it-,
90
2i o
C
510
F>
C/cc k p
16o
pF
0
#<o
't0k
2- k<
the -3
dB corner frequency.
For
desirable when high frequency clocks are required.
Multivibrator" was used.
The resistor R on the clock output
to protect the Reticon filter when the supplies
clock signal
is
off and the
are
is on.
"Peak Indicator",
II.
these
Monostable
the SN74123n--"Dual Retriggerable
reasons,
and
implement,
A duty cycle of 50% was easy to
adequate.
is
A TTL level voltage swing
Detector
or AC Voltage Threshold
(diagram A3)
This device
voltage has
time.
The
been reached
and maintained for
and rectified
is stored on the
transistor
amplifier
/
be rectified,
wave will
LED indicator.
indicator.
resistor
In
is exceeded,
stored,
trigger.
to
to
The rectified
and measured with a
operation,
the
if
the
output rectangular
and used to drive the
It can be made more sensitive by
in the Schmitt
applied
and
(with a 100 ohm resistor
1 uF capacitor,
Schmitt-Trigger's threshold
of
The Schmitt-trigger
protect the diode from high current surges).
voltage
length
a short
attenuated,
(with '4V hysteresis).
AC coupled,
is
indicate when a selected AC
to visually
is selectively
input signal
a Schmitt-Trigger
output
is used
lowering
visual
the 2 kohm
IIL'-
iog
V
0-,
#4-
'IL
2Io
AAA1k
----
7plee
cVaCT; %-eSb,/c/ Zpe/kc o-
Pre--Amp
III.
diagram A5 for
allow easy
coupled to
Hz).
an
This
input
the calibration
AC amplifier
audio taper
is followed by
and the high
system,
(see
illuminating.
active HPF
to
and to
devices,
It
is DC
(gain = 76) with gain down 3 dB at
160
(down 3 dB at 200
a 2-pole butterworth
act to set
pass band points
not
attenuator
input gain control,
curve) for
a 2-pole butterworth
These filters
frequency
at the
on the Pre-Amp schematic.
interfacing with various external
is
Next
6 kHz).
band,
an
so that the distortion LED is
adjust
is normally taken
(diagram A4)
amplifier has
This
device
low-pass filter
of the 2-pole
output
Hz.
for this
input signal
The
LPF,
the system's high
(the lower frequency
down 3 dB at
and low
edge of the first
frequency edge of the 3'd band).
the other edges of the 3-formant bands will
Later
in the
be defined by
Reticon switched-capacitor filters.
To control the
amplitude
attenuators that feed
channel's.
control.
in
each band are three audio-taper
the pre-amps's
output signal to the three
See diagram A5 for the calibration
curve of the
) 6K
Pao IE
q
O J rol-m- A
.000
1 Z??pL/
O-lok Avdio 0Th er
V
V
c
C)AN41EL
I
74-e
-
30
p2s2o
'2o3
Li
Ztz:
~
1/0
Is-
5
35'
b/
30
A5
t
30 f A,
3Sfia-ortCx
s
7
)PI5
/0
k-
5
/9
2e
36
4/0
re
t
7c1
.
O
/t'
Pi/ ge4+h-
/
* 2.6 36 I/6
5o * 0
0l
68
Pre-amp signal
A6)
(diagram
i'st Channel
IV.
'a'
input to a Reticon LPF,
is
the higher-frequency edge of the
the
"First Formant Output",
anti-clock residue filter
amplitude
frequency
vs.
which
AC amplifier.
eliminate
the hysteresis
frequency curve,
is shown
relative
With
The
whose
amplitude
input of the total device,
in diagram A8.
output
anti-aliasing LPF before being filtered
by
is passed thru
a Reticon LPF,
remove the higher harmonic distortion products of the
to the
to totally
is 70 mV.
is test point two,
The rectangular Schmitt trigger
waves.
= 12)
DC coupled to the
is
"dead zone"
to the
DC
(gain
(I would recommend AC coupling here,
output of the Schmitt trigger
Vs.
the
the DC output offset of the AC amplifier).
supplies,
nominal
output
and
the
test point,
The R5609 also has
is removed by the AC coupling,
trigger.
Schmitt
See diagram A7 for
response curve of this
The amplifier's
1,
Test point
at the output of the
=1234 Hz).
input of the device.
relative to the
offset,
is taken
(fn
formant.
first
which defines
Test point 3's amplitude vs.
input of the total device
an
to
rectangular
frequency response relative
is given
in diagram A9.
The
post- IPC filtering of distortion products below 200 Hz was not
considered necessary because of their relatively
and
low contribution to
articulation.
low amplitude,
pC) e
00
..
_x
lbek
1-
)0k
3
10k7110
.T
+10
T-P 3
5
DA
c lock A
1)6
7
I ))+
--/0
C
a72
7e(
I1
KEUFFEL & ESSER CO.
18 X 25 CM.
46 1523
)
MADE IN U.S.A.
4
jji
1 7,4-
L-LF
_L
,,
Lrt
,I
1-111
+_L
+
17
+fl
L;_
-
j
V
t-
F1
_Aii i
L1 Lj
I
_T
4-14
LLLL14LTt -1-,
L
P
T Tl_
F
7
i-t
Tk
-1-
F
+L,
_F
-I
441
4
T IN
-rrr
r
t
4-4 1
- --
j
I J 14 tft
44- -1-4_t__r
--
F,
-
T
_11_:
+
L
T
-R-rr
T
4-It
__
#
+
-
11F _ _ _ I- --1-1- ti
l-,
r-l-l-FIr
_:F
4H
7
r
V
--- tt
_jL
L
L
4-L
I
1_
4-t- -
L! _L
L
I
IT r
+
L
;r-j-
4--"
-1 _T
Tr_
L
7
1
L
7--
0
T1
L
F
7
F
A
....
it
#
__ 11
L
7-T7,
1
1-
1
ti-
I
1-1-t-
TIJ
-
X,
-1-1 7_7
-1
T Fj
I _tj---
1 1 -1
7 1
T-1
4
ri
r
H
7
T_ T717-[
4:-
t
_ILL I It
L J E
1
rrrr
_!
_Ltil
L
11-11- -iHltllll
7
77
T
F,
-1=1 p
I------
----
i:
HTT1- -q
T
-F
_T_
LLL
t LLL
77 1-
r
T
LT_
-1
I _r T
FT 7-
I
+T
+
Hj-
LL_LILUL
Ek
-Lt
I- -I
I
r
4_1 !,I I
WrF"
1411
L
-11
-1-
- -
7T
T
-
-
t
-t
- "I
T-1
4H
7
ILL
1:l pt - 7--
I
__
-T i-
4
4q
L
U-1,
L
tt4 L T
It
F
_F
gi-LIT
ItEIIL
I -1
I
r
rt
L
-i-
I
L-L
LiL
-A
4
1
T
_rt__
it
LH
IT
L
r
m
IL
l
I
rr
+1f-
11F
E
T
I,
I
TR
Ti
rT_
1
If I pvl:
-14
+
17
44
L
IL
-1
#
7-:1-r _ITTTT
I
_r
IL
------
I I
+44 14
It
7I-_Tr1
Ell
Alf
1:,
H
TTir
1
-
r _tp th
J. _41-
-
EL
-I-
F-H-
It-t 4-L
I H-1-
_LILL -- LI I- I LA I
!
It
li, 1"
4t
+rF-1t-_ L
it
I L
A
j4
1 1
R:
:4--
4tt IT,
L
t
1
44_1
I t+
I h,,
EL
#
T-1
J
E1+44
-T
T
t
L-I
1 1 1
---- IT , I I
4-"
N;;
1
L
+ -- -
1 1,
T++
+
Lf
TFITT
I
-
LL
IE
+
L
T
+
1
10OX 10 TO THE CENTIMETER
L
10~2
I
lX 10TO THE CENTIMETER
KEUFFEL & ESSER CO.
18 X25 CM.
I s5
-_4
fi H
4
_ji i
_LtH
LL
-T-, -1 IA -
Tt
L
I -i
LL
M-
-U +- - iLL
Lj
I 'r -T- J-1-1-ii
,
i:,
tF
LJ_
-L-L-
LT-
-4
L
7__ ri_ _:
41-
-4+
1-1-t i
P
L
I
A
ii
T1
L 11
it
FFI
4-
Fr
I
F
L
F
IF
rt4
TF
L
-f-
+I
it
11 T, 1-
Tt
+
+i
LL
T'L _E
+
LL
P>C
i- r-*io
+
J,L
46 1523
MANSN U.S.A.
Tu
4t
114
4
_r
L
_F
TT
E
+
it
47
F
77
I
LT -E-
F -r-T
+
z F
r
--f+'
I
_j_L
t
T
T74
TF
E41 4
_A'HJIT
1_ __Lj
-Ir
k1- -Ir
i
:L
- iL
T7-1
I
IF
4
LF
L-L
L
1_1
It
+ F4H
T
I IFJ
F-1
:,-
I _ Lt
I
FEE
r
:F
J-1
..
. .....
11
4,
mw_
Fr T
A
7-
-------
_7
I
m
-1
L
7.
T
-I- i
4 I-Ft
L
J4r
41-
---1- -
t-
t.+14-t
L
FF
L144 1
r
IF I 1 1
71 H- 7771,
1j _LL-F
71
T,
,41-1
#
+
-i L
._JLE L -
JU
-'t-
_14
LL 6-
L
:1 T
+
jjo _L .
Al!t
'4 '
---------
L
1 1-1 1 1T
ON
Tv
17
All
-Lt- L_i4
I+j- +j
_lA_
F7
-F 4-
Ill II till
71
Ff i
Lk +
0
+
IT+l
1tz 111
-0 1
-1-1 F-1-
r
+E
4-1- +flL[F __ t _L_LL
1
1 I- -I-T
l
L
1 Tj 1
__ __
1444
ttriti
4-
I- -I
7
_JJ
+
-T.
4-
_Lr_
am
i7;,_1 4__
=71=
r
J'
AI
4-
R
til-:-F -
T
T-
TLLL
LFl_,
+
J_
J-1
_T
44
7114
+ +1
Fl-
4
4 -,
J!
ZIt
T__
. . ...
L
-Fit -iI-
#:
I- rrrr
I
-
L
L
r
- ---------
+
L
------
-
LL
i'l_! IA,:_A
-LL-L
L
qit
tt
+
T7
I -it
i44- i
I
.
I
-
-i
-Lt-
+Ht -1
-
__T
_T LI-H- _L
Lt
4
A-1
1_Llk-
-------------it 1
FIT
,[" T
ti
T
T 7_ !!i
4-L -
- I
fF 14+1 FH-
- - I _11
k
-1-1t i-f
H1 T
H
1Ir
r
r
-4! I !J4
-I
t
H_
77-
_LLLL
P#
i,
1T
Lt
Ir
4
_T7 i
iT
+ l,
FIT-
7
1A,
_LL_
Frr
+
+ 4
TL
+
_1
A
7T7
T_
q
+
+r,
-FT
I
L
+
+
L
4,
4+
4
Ti
TH
-4-
mLLLL -
rid
LL
_j
J,
4 HA_
it
A
D,
t]
ir
18 X25 CM.
THE CENTIMETER
10XO0TO
& ESSER CO. MADE IN U.S.A&
ll' lox
KEUF'FEL
46 1523
A
Ll 1-1
20 (ff_ IV
+
FIT
-1ji-
T
A
L
r
ip
t
IL
1-1-1
ITT
_L
_: -
pH
,I
1-l
-1!
ti
EL -EM-
Irr
-4 t
r FT- I
LL
r
t
r r[IT!T
E
jidli
T
I
i
if
F -
H
41
I
-1
" -1,
1
_LLL L 4
4 i 1
I
71
1+
I I
TIT
-
T
U
it
-ta- TF IT
T+
Trq
TV
Ll
- --
IF
--
-
4
A:
+
di
t
H 1i Lit,
_j
LLL
+
L EF,
+
I L
-rT-
-Hit
.1 IT
-1-
77-
1-1-17-1-
_LF ItH
MT
F4 I
jw
I
-f-t
yj I
_T
_LL 11
44
T-14[j-, 11:
4
+4LL
111
i4_4
-T ttt
4F
77
it
TIF'
+
r
_T
WT
L
t- A L
11
1_7 1, 1-1-
I
IT
- - ------r
I
-
-T--H
4
-F4
+
T
Tr__
_T t,3+t
l -t t-t- I -f
F
_1
4
_fz
Im'
ZLL
I+ f
JL
11
Iff
#
k
1
7
_H:
TIT
TELI
H
F T1
4
- - - - -
:F=,F I___T_
_T_
_rt
Fi
-1
_rt
t
L
i-I-I
ir
I1L _k - _."__j
t7+
LLL L
t-
i-
IL
-
-t+
#
I
- ,L __
-r
ItI
-14-1
Iii
I
-11-1 j-tFF fqTP
t i
-t
ME
TIT
T
ti
H
4-
-rr r
T
L
it
4-
-t
it
Tj -r
Fir
- - -14
-Tt-1
-_El
-4
-
a
-4
1i i
L
:Lftt
OL
it
I
_H
111-14H TT
!-1- -114111 - Ali
lit!
-r -t 1 4
1
-1
+
I
_"
4
117
-t- -
4
4-1-
L
_rrlA-!
r
------
4+
1
Y1
IF
I
L
_++
-tj
Tl-
-L L
J##Tq
_l1
+
4
f
f I
;73 Elr
-1-i-
-rjj -1-
it-
4'
7- - i-
T_
II
_rT
I __L
#
F
1
1144
44
+i
Ti_=
E
_1 T -1-1 _4
wj
fi
Jill
it
t
: 1
f 1-11- - -,
__E -1
;
..
-1 T_
4,
401
L
io
L
11
I
I Ll I L j I
L
'I
I
T
t
,r
q
47
+
1- TP-E 1:
q
+
WF
.
1_1
1
-
4
4
+
7
+
i,
+
r
2'nd Channel
V.
(diagram
AI)
This band has Reticon high-,
The AC amplifier
frequency of 3700 Hz.
natural
identical
is
circuitry
to that of
low-pass filters are used on the
higher
harmonics
"2-'nd
Their
to the pre-amp's
relative
1.
channel
Schmitt trigger
/
Reticon high-,
and
IPC'ed output signal to remove
Test points 4,
formant Peak Clipped",
signals.
PC/filtered"
set the
and distortion products that are outside of the
second-formant frequency range.
"2'nd formant",
to
filters
low-pass
Anti-clock residue filters have a
2'nd formant.
range of the
and
and
5,
and 6 are the
"2'nd Formant
frequency curves
amplitude vs.
are given on diagrams All,
input
A12,
and A13.
3'd Channel
VI.
An
(diagram
input Reticon HPF
edge of the 3'd Formant.
circuitry
the
caused
is used to define the
The AC amplifier
identical to that of channel
is
IPC-er
A14)
is used to prevent
by the
IPC-er,
frequency range.
8,
i.
Schmitt trigger
A Reticon HPF after
low-frequency distortion products,
from appearing
Higher harmonics
a single-pole RC anti-clock filter
single-pole
/
lower-frequency
anti-aliasing filter
in the second-formant
(above 6 kHZ)
(fN =
are removed with
7400 Hz),
(fN = 7400 Hz).
and
the
Test points 7,
and 9 represent the 3'd formant, 3'd formant IPC-ed, and 3'd
formant
IPC
/
filtered.
Their
amplitude vs.
frequency curves,
+1I
-0
L0
77
.7
~~~~a
)Oek -g
-)a
uI
j7/yhwre
U
0
o
ow%
ILdJ
10
=S-
4-4
14-
4
I
I
H4
L-E-
L
+
t-T
J-
L
tlTH
--
-7
=T
J-H- -H-
rn,
-T4
---
J-i J-L
4
-------
Tt
T
#
TT7T
7
--
-7
I I 1 1-41
L
11
T"L;Ij
LILL
LE
-J
Ai
4-1
I tLl
-1
fi
I rT
T
+L
117
Ft
t+t
Tr-r
L:
-1
+
+L
L
I
+-T
L
-4
4
-
rT-44
r-
Ili
Ii
ff 44
T-
T=
v
4-
T
L
,
T------
E
I+
7f-
Hi
7,
11
_I-)
L
1
4
T
j
JF
1
I-E . A
L
1+
T 77
I tj
ILi
I
L
L'- L- -
l
1
I
Ht
i- i l7
4
f-T
Lit-
1
L
4-
I I
7-
L
T,
r
1 t-r+
I-L
ft
F
-
.
-A
!
+
H
..
'
17
-ED
11
4;
-177L
L
IL
_LT-
- i-
.-,
F!
H
-
-LL
T
-L L
7--L--LL
T-
T
i 1
T-
-w-!-i---it ++-
-1- -1 =
L I-LI +
L
l, Er
14
T7.
_,'LL
r-1-
-1 L 7:
LI- it
77-
+
I IT
I I
T, I
-I
+
-4
LL
-LLt
_LL-
TFT
tt tt IHH,
I [-J-L -
LE!
Ti
-1-7
I L
L
LL
r-
7 1
-4
4
-l
1 1
JL---I-
H! +1H
''1
4-
L+
T4
LI
+
JI-1.
-4-
A-LLL-
-I
1
-LLL'L -i-ml--L -jif
L -1
, + +H+
-1-1 L!
+
4-
7 -1
-L-E
-I-
L +-4
A
1 IT L
#L
-T
4
I
4L
-r-t
I
7
T -717
-L
+
--- TT-T- -1-7-;
Et
47-
-
_L
L
L -
+:
A- -
--- -IT
IL
MM
It
HE
T
-PIEI-,
I Lr
, 1
IF -T I
-T-It
F+ I
l
7
A
j
it;
Tti
H14
-- i-L
14
LL
r
t
1 .
-L
OOFF-+- -- A-
it
-EiLHL
LLL__
-P-
-
OWL,
,
7
4
77
+
L
f
L!-I-i-
TH
L
1
am
J4 H
.L
-L--
7
E
11
L
1
I
I
L
1 T
-1
4J-
IL I
A
-i-7:
T
i i #4
L
4
17 EP-T
i 14
I
A 1 4F
1-,F
-J
-4-4-4- L--L4- _1
7,
-EL
II-TT
+j
+
-E-4,
4
TT
II
-1 J
+
H
TI
--j+,'Li+'
L -t-
L
1,4
-H!
14Q
311,11,
]
I
:L
I
I
-7
10OX
10 TOTHE CENTIMETER
[~'~KEUFFEL & ESSER CO. MUDE IN U.S.A.
2)"lof4
~FLI
L~
-TKH-H7F~tT~1V
Fo tm
46 1523
4d
p-c.
-1
t~4+
FVFYI
-d
Aj
F4-
4-4~r
4
-I L+4-
18 X25CM.
ti
IVI
j:, -LJL
4 -- I
4n-
FI-
~t** ~
1
H-'*
-~
__
Tn-
H-
F&
1 1
Li
~
-
-
"--- - 111
H d
t-
-- -i - - -fTF
- - F
-
i-
rt
211
H-
tt
ttr
I
1KF-K
-F
-I-
HI
0X
KEFE
18 X 25 CM
0T THE CENTIMETER
& ESSER CO. MADE IN US.A.
-Lf-l-.4L-!-Tfl
-
f
46 1523
P
Fv s- -m AJ6
KEUFnELd
L7
L~ -L
-r-
-1-1
~ L
4+4h2
4TiI
#*LI
L
444 +IL
t
4
it-I11f j- ''4 14i
L 7
rVm
F1~~4
I
1L
F
+I
J ,
-ILI
'
-t
I-
T
14-
+
7LLrT
-LEF
1,-
41
H
7h
FIVE Tri4i~~
Fii
-tf-
41
FL
#
-L
ILA
4LLt
LL
<
v1
LL4
-- F- LI
T
4:
4
FL
Lir
rl LLLI4
L
F 4 Lz
fLf4I
h
<
K
L
F<
L
1-
-r
itF
I
T+
4
-E~
-
I~~L
[7L4LIT
4
4-H
LLLI
---
12117+
L
IL
44LIf
#4
-44
LL
IL
*L
L
4-
ifIV
LL[
L
4LF
L
1
I
V
L
~
7
1
44L
7
_
~~2
<12
f~
i-
f rr ,tH
-
41
-ALI
1~i7'
T
L
-~4
LL
4
~
IF-
I
-
If"-IILT4
-
17
IOK
005*
cloc- D-
, cz)
15(1
I-0___
c/ock-t
<e7
- id
cZYne/
input
to the pre-amp's
relative
are given on
diagrams A15,
A16,
and A17.
VII Summer
(diagram A18)
This circuit adds the signal s from the three bands,
divides the sum by 3 (to
amplitude vs.
The
three-band-clipper
(output/input)
is given
in
The
diagram A19.
app arently due to out-of-phase
in the crossover regions
addition
The overall
downward slope
anti-clock filters
by raising
(at
be used
Recommendations to
AC coupling
Or,
an
instead, with
This
active filter
a higher fN.
improve circuit:
As previously mentioned
amplifier
to
the expense of greater clock noise)
the fN of each filter.
(multi-pole) could
channel,
in the spectrum is due
(single pole) on the Reticon's output.
effect could be reduced
1)
for the complete
frequency curve
900 and 2800 Hz are
at
"dips"
in the op--amp summer).
clipping
prevent
and
in the discussion of the
is recommended between the
and peak cl ipper.
IPC-er's
I'st
lX
10'~A
10 TO THE CENTIMETER
18 X250CM.
46 1523
R~~~KEUFFEL & ESSER CO. MADE IN U.S.A.
7T~~~~~~T
F_
I L
+1
77
_TYV7FJR
4
For4t
-3
41ri
JA
_L!
14 -
_L
144
E
T
-M
q--
7Itit--
-' A
L
FEi
7
1
+
:L
I
I
,
I
-,-_____
__
ff I-..
17
L
T
Iih~ -.
I
L
FF -'
-
I~~~
i
______
4'4
iL
FE 77
4i
-
t
F
I{-
L
-
_r
t
T_
AT
I<_4,
X__
-H
I
Fl
=1
#
L
_J
IF
18 X 25 CM.
1X10TO THE CENTIMETER
CO. MADE IN U S.
IK,
46 1523
KEUFFEL & ESSER
_Ij
Tt -i 1
-T 11111
4
::fT
ER
14
1-4
t4RITr '41
I TL
A_
A_
1 - 7-
11
T T'If
r
L
4-
+
J4Hi
Ii
41 Ht
,
H-I- -J-1-1-
I- E-H
El
__LL
I 1_4
t#
FE
P
t -1 TH FF1EL
1
=1
I
TF-
I - 1-fi-4
44
T-Ii I r
1
--t L
T fRr
H+f- -1 J
filIt V
"
4
IM,
-T
r
T
r
T-ri
F
0
F I
-It1
I
Id
FF
41 H
I
4
14-
i
I
--
_#H
,t Ill:
1A
i
1. +1
_7r
_17
lrl-
L:
I
1
-1I IT
JI
Tit
14-
il _FF
I
-
-U+
LL
F
4 i ,I
I ---
je
I
LL
T
zL4+
1
7 ...
tjl
L L
T
- - -------1411 4 t-,,
it
1
LL
I
7
H
_L 11-LIF
T7r4 1ij ,
T_
If
11--1-1LLI_
i 111 1-T
t L_
4-1 4:1-1
H
&LE
-
T
1-31 TH
7
d
4
#it
+4-tt
W
T
-4
4
1 1+'
T
1::
TIF
TE
T
IT
j I
i
L-L
L
114- 4
1
I r-! r
EF
r___ I-
7 _T
F --i
IL
T
T4 1-[
111 mr II iii,
:L
-TT7
44
-1 4-
F
I
II
4
1
F#
4
_T1
+ -1
T_
-1
Al
W1
aldl
-,41 1 ..
7 "IS-III
I
T
tTq,
LL1+
-i-
-
I
-
14W4- -I-!rr
4__
- ____
++
L
L
ri
II I
I i_
TF
I
__
4
Li-L
4 i
-
I
71-
__ '+- __ __
T
HF 1
-ITT
i-
11 .
- - -It
:pp
-
l4-
LLIF
-----
+
I
---L
HA f
-
-
_14- -
T-4-----
-
1, I I ; -r-L1-f
71-
-
1L "H"IT-H-H-
+
f
I L
II
4
II T
4-
Ir I
If
+
I
or s,
-It
-
tL6
Ll
itt
a
M
---
+f
;-L
-t-
!1
IT
T
I
T -i
IF
jLLL_
i 1
IF
T!
4 -1 '17 ,
T,
7-k-1
t
__F
E-1-_7
:4
_j-
-
E t
-T
_LA_
I
-L
L.LL __F_1_
4 -LL1TF T'_
mf
..
14
id fil
I'd
Tf
f
L4
-I L
f 1 1,4 ---L
T:
Tff
T
IT
PIR
-I
.0
L
T
+
E li- -I
_L_
I ITT
T
T
-- ---;-74 - -----
+
T-
TF
:4
_T_
F-
L
7-
-F
[4
L
-M
4-
L
-1
F
T,L
1-1,
- T ITI
IT
I It
F
-4-44
-
I'll
-I-
Lt
-
-F-IL1_4-L I
L --I
F-7
tfT
J-1-
-
J A-
>I
-1
18 X25 CM.
OX 10 TOTHE CENTIMETER
10-'
KEUFFEL & ESSER CO. MADE IN U SA.
46 1523
42 o coa V
Li-
L-
-A
r
Ii- I I-
I
11-1
I-t
LL L
LL
I-
IL II
4
m
t4-1-t
I r
J_ -
,Rl
+-FF
T
I-at
t
7
- -
+
IU-
t
+
J j_' : 'I I F
-, J
tt
t
T
_tL
17 +1
+4
4 1
Ll
4-
4-
+11
-4
Ill
+,L
4-
114+
444+4+4+
+
TT
: 1 wL
T
I
t-
L
_7-1-1F-_
t
T - -V
I--
-11
F 11
-fit
1 r
#
I
+ fll
-
J
-
LL
+
I1
___L
-- 7FT_
IF
I
H+4+'+-T71+-i+
+
IL ll
II
rr
7-4-
t:
7q,
tL_
LIT
-
-L, 1-
ft -
1"
L
onso
:L
it
T
7
-F
-LLL--
I LL
H-1
-14
"P , W
1-1-1441
If
-- ------ ---4THTR!4V
TF
4-
-A-1
LL
id
L
TT
T7
+
--ttL 11-L
+
J,
I-
4
TII
L
4
Ll."
L 7TL'TW
J,
_T_
L
jt_F I
IL
!#
I
-4T
'H Fi:
I-Tl-f-i-4
-!4
IL
1
--T-r
ii4
14-1
L
7
t-
I- t T
TJ-7
+
-T
-
:!7
4
4
1_1
T=ff
7;
_J4
I
F
L
]-I I+
LL
-1
L i
IJ
'L
It
T
Pf
H+f-f
-I
-j T IH-
L
I''LL
I- LH
+
r41
f+
1 41RM
F
ILL
t11t
---
-4 - -----I
-----------
+
T
J_
L
L
-7-
11+
FAN
+
r
H
r- 1 t
-
[1
"A
-W
AFN
41
V-
-4
4-L L
+
7
+
L
i
:E
L
4
L -
-
-1---
7
t
T'
F
AL
L
14
-r-
I RIM 1
1f-
th
+ 41+1
4,
it
- -41
T
t
_T+
-I t t-I
__7
L
+4-j+Fl
-
U
_j
T'LL
lFFF
-1
, FF
1
W
-r- tr rt-
Tr
-r- _' I
T
:t
I
-T
A
I4=
r
1f
T
F
fil
-
F1
J+ 4 11
o _i_
-
,t
I
Al
_F :JL
TI
Fri". - 4
-
-j 1
+4
-L-L 1-
-1 A-L I- - -L
7
-I- I
_t T-I
U
I+
L
I
I
it
-444+
T7
!4-j
I
T
tt
4-
T
t
44'.
FF
t +1L -L'_
f-4
T
T
_T i
I
lip
+
L
-7,7
F.
L
11
--
1-
i-
u~
-4-
(*
4
_,_j
4-
L
t 7-t
v
#
ir
H
_L,,+
4
i
L
7
T
L
4-
-4
IAL 4
_4
-4,
I-!-
I T-,,
FE
P144
.-A
Ld
J
J_ _ 7__zr
_J4
HE
L
TT
LTL
-4-
L
H
il_
J
4
4
L
L
4-1,
IF
7
TTFE
7_
7
-4-
-7-
_4'_ 414'1
4
tt-
+
t
_rF, I-
+
14
--i--
-
_rT__
jf
H I
L-LL
#
01:
JIT"i
4
-T,
FT
"I
t-J+
A-
i
1-1-
41
L
4
LF
J_
_rTI
-fT-
E
H-1 I
HIF
T
4
7 _LT- -
A
7 1
_T
4-1
I
+
--
L
7_T_
_7
_A
- . ..
II
-
I-L
p
A
.....
ri -1,
-4-
r
T-t-
+
LLLL
_LL
i LLL
7_17
TjX
Ti-E
TT
..
.
.
7
E
T
j,
T
-1-
I T-
L-L-L
T-
I
t=
L LL
-H
+
L
1
F II I
TI
L
4- -
7
7
77
77
7-1
+
-- - - -
tit
I-L-,
-A-1
1- -4-1 1
-
J -L-L
l_77 -----
JL I
LL
-TI-
J
7
4
JL
-
__L
+
L
T
L i-I
Lj
7
+
.....
.....
FI
t 7 ---
7
77
L
f
"IE
:
j__JL_ -
f
1_3 LL
TT
T I I.
LI
LL __LL
__LL1_
7
__7
I
I L
LLI
I E HE L-1
1
it
TEEE I "
1-L
T.717
r
I
- _+LL
I+
LL
-1-
_L
L
L jT
T
41
ilf- T
17
7-71
:h
4
-L44- j:4-,
7
J
1-1
7F7i
-1
_J_:
-1
_T-1
1 IF
.77
4-
""I
:1]7
-1, LII
+7-
TH
+
X
+
A/b'
56o p
kok
/
cZ
AMA
}O6&
AAA~
/
cEID
~
F-]
~3J~
S u722279fr
A 18
pli7
IE!
18 X 25 CM.
1X10TO THE CENTIMETER
EFEL & ESSER CO. MADE IN U.A.
Ir
II
-I
F
I
--
-
I-
ILA
_Lj
JR
4
1A
....
.
T L
OVrrPLV17_
I
_f_I r 7
r J Fir
IL
46 1523
t
1 4,
""
1,
,
,
.
-i
I
7
-T
4
T
-1 i
_T T-l'
L
L
-T -FF
T
Fi-
AL
+
4
LI- j ,
It
__ - _L
- --- ---
- ,
-
-
i
*
+
T
777
-i
+H+
+ ++
__-L-M
T
IL
I
it - 1
I
L
FE
FP-I
11
L
t
F
IT
_t_
4
T-1,
T-
_TT
- -
I
I I'll L_
TI
- -4+-[-. ;
H
I-
I Ill,
1
I
t
ip
F
HV -+i
I-
_T
L
K
J-,
[
t I
L
-,++-i
+1
ALAI
L
.. 114 1
I
j
7
tq L
T-F
_T
i-
-t
1--i,
,
1
1 j,
:-I T_
F1
IT
+
-44
IT I'
tL ,
F
II
II
_7 1 1
I+ -
LL-H4
L
It
-1144- h
i4_
Lt J 1
J -90!p
A
4
t
I -
r Tri
[41-1- 1
L
T
4111
-H+
7
L
TrL 1,- 1 t
L
-1Ll -I-!
_LL
T-t
#
I+
4
L+_
L
I+ _FT_1_r T
1_
-4
IF
7
t 11
---t-IT-H-v, I
4 -11 41; 1;!
-
F
FF_ TI
ill11 7 - _K
iI i I
1
11_
.I
-
1-
1
7
f
4
-I---
T
4
+
Ei_
t
IL
Lj
-L-L
-
t
ti
-,-R r-,r
IIr
i-L
TP
+
Lt
L_
IITL
-1-
MIT 7'
+
H-1
i-uJ:
I
ti
-
:Z
Xi-
:I
it
-it,
_4
T_
ILL _HL,
...
Li
r
_T
-
L
_L_
-7
TT-
+
+
ITJ:
I+
v
_ttt
_T_
F -1
-----
Irt
UL
Tri
i-
r
4
I ---- --- T-11
L
ti
-1 t
TF
+
-4L
L
L
AL
_F , '111
-,Fir FL_F T-IFF,
L 1-1'1
T
E
-
IAlt-
H-
,
F,-
+
TLI
t
Ti
...
t- -
-L
A t
IL
L -F
L
T
IM
LL;
_L
A- _ki i
4-
_-'l
U
Ft
L
WIT
_L
LL
_71
_T
A+
I
_t
-- T-11 r
LJA
41
LI
,T
IV:
I 1
did
_T
+_T1
1+
J_
I
T,
T
HHI
T
i
t
it
It
f TL
f
4
-1,'- 4
1:i- L
+
tl
i__
r
iT,
!41-11
i
L
L
+L +FL-1-
#
L
LL _L
-.r
IT 44,
-F
A J
-FIL
Al-L
L-1 It 1
__L L -fl +4
I Lj:[
--i-11
i H 1-1
i__l
LI-F- _144-1- L_
-I _, - --- -- -4. ----
F
H+
L
1-Ij
L
H
H- 1-
H_
-- 4"'d
-
r
i
41[ -44 14 +1
4_1
_t , lti
-
1 ;-1 1 L
-b-
77
1
L_
I+
44-L, jj
4-
4-1-1-'-Ij 4-4
_1 1j
4-
-L
Irl L
4
V pF T;L
L-1-4
_17
LH
F
---- ---- --
7
7
!
t
T
I
rL L
f-r
L
-1-1
4,
HE -
1
rr
+
-r1:1-1: 4
1+
T_
2)
the
"R"
The value of
clock schematic
feed-out resistor)
(the clock signal
lowered to make the
(diagram ) should be
shaped
on
opposed to exponential).
(as
clock signal
more rectangular
The resistor
is present to protect the Reticon filter from
think that these
to the
nulling of the op
(output voltage)
input dynamic range.
Peak Clipper"
"Infinite
Niederjohn's
Also
greater
"optimal
filter"
"optimal filter".
sensitivity control
hysteresis
and
B.
I.
Thomas
J.
and R.
(diagram A20)
constructed was a battery-operated Infinite Peak
to be used separately,
it
amps
and
IPC-er could be done to obtain smaller hysteresis,
VIII.
before
protection
filters.
consequently,
Clipper
I
at the output.
give some
and still
lowered to '400 ohms
3) External offset
in the
9 volt
causing the spurious
are
(high-valued) resistors
and whistles sometimes present
clicks,
They could be
and minus
are turned-off before the clock power supply.
power supplies
tones,
plus 9
if the
damaging current which would occur
is
The
IPC has
or
in
conjunction with Thomas's
a 10 kohm
to vary the gain
made rectangular by
"dead zone" of 88 mV.
from each 9-V battery,
quiescent,
into a short circuit.
Output
impedance,
input
of the signal
(0 to
and
a
151)
a Schmitt trigger with a
Current drain
to
about
impedance
is
11
is
about 3 mA
mA drain operating
1 kohm.
9
0
j
/0P+ 17n 4
i::
(use
2
A
-
17
®7p,
Ok
In/x
lo /0K
i
Lo'T,
)0/<0
LA/t-noIe
~lot<
56/r
k4 Cp
~42~]
87
Aoe zc)
8
The "optimal filter"
frequency of
It
i100 Hz,
is used by plugging
and
it
is
a 2-pole Gaussian HPF with a -3
an asymptotic slope of
into the
provides the necessary 10 kohm
input
impedance of
of this filter
500 ohms.
input of the
load.
The
2 dB/octave.
IPC-er,
The optimal filter
amplitude vs.
is given on diagram A21.
dB
which
has an
frequency curve
4652
18 X 25CM
10 TO THE CENTIMETER
KEUFFEL & ESSER CO. MAD IN lU46A52
W' 10OX
r
I~fL
F
i-I~i
Tl_I
I
1
fI Ll~VI4~~
1111~~~~~~1
i- 4_ l'7
I
A tlf
I4 -
I
TLI VhLL
THT
1 44
1 14F H~
1i
--41
rIH
, -- r r- I A-- - ---
1
i
_4
41
IE
#Lh L.-H
-t
i 4
IK
1L
I1 I- __LV_ 1,---T_
-
JH~
I
1L
I
t
1+ -V
11 2
...
I
...
-~I
I
I
1HA11j1I 1_
L+II
V>
L
1L
I-TNI
14<
~
..41~
I
41{
_T_~fI~
I
REFERENCES
J.
EGAN,
58,
Articulation testing methods.
H.,
Speech and
Laryngoscope,
(1948).
FLETCHER,
Nostrand Co.,
(1953).
for speech quality measurements.
Audio Electroacoust., AU-17,
IEEE Trans.
Van
D.
in communication,
hearing
Recommended practice
IEEE,
LICKLIDER,
the
P.,
J.
C.
amplitude distortion upon
Effects of
R.,
intelligibility of speech.
J.
(1969).
Soc.
Acoust.
18,
Am.,
(1946).
C.
LICKLIDER, J.
R.,
BINDRA,
D.,
and POLLACK,
intelligibility of rectangular speech-waves.
Psychology,
LIM,
61,
J.
Am.
I.,
The
J.
(1948).
S.,
and OPPENHEIM,
A.
noisy speech.
bandwidth compression of
Enhancement
V.,
Proc.
and
of the IELE,
(1979).
MARTIN,
S.,
E.
and PICKETT,
transition discrimination
Acoust.
Soc.
POLLACK,
Am.,
I.,
distortion on the
Acoust.
Soc.
48,
J.
M.,
F1 masking of F2
in hearing-impaired
listeners.
(1970).
On the effect of frequency and
intelligibility of speech
Am.,
24,
(1952).
amplitude
in noise.
J.
J.
67,
I.,
POLLACK,
peak-clipped speech
31,
Am.,
at high noise
Intelligibility
B.,
I.
The
formants
and second
influence of first
intelligibility of clipped speech.,
on the
Soc.
Acoust.
J.
levels.
of
(1959).
THOMAS,
16,
M.,
J.
and PICKETT,
Aud.
J.
Soc.,
Eng.
(1968).
B.,
I.
THOMAS,
and NIEDERJOHN, R.
intelligibility at high
speech
clipping.,
Aud.
J.
THOMAS,
B.,
I.
Eng.
noise
Soc.,
16,
J.,
levels by filtering
and
(1968).
J., The intelligibility
and NIEDERJOHN, R.
of filtered-clipped speech
Enhancement of
Aud.
in noise, J.
Eng.
Soc.,
18,
(1970).
THOMAS,
B.,
I.
and SPARKS,
D.
Discrimination
W.,
filtered/clipped speech by hearing-impaired subjects,
Soc.
Am.,
49,
THOMAS,
Soc.,
THOMAS,
Aud.
B.,
and RAVINDRAN,
A.,
Intelligibility
already noisy speech signals.,
J.
Aud.
Eng.
(1974).
22,
spectral
Acoust.
J.
(1971).
I.
enhancement of
of
I.
B.,
and PFANNEBECKER,
weighting of speech
Eng.
Soc.,
22,
(1974).
G.
B.,
Effects of
in hearing-impaired subjects.
J.
WATHEN-DUNN,
clipping
in the
YOUNG,
whitening
L.
W.,
W.,
band.,
J.
Acoust.
GOODMAN,
J.
T.,
audio
L.,
and LIPKE, D.
and peak-clipping on speech
presence of a competing message.
On the power gained by
Soc.
Am.,
and CARHART,
R.,
intelligibility
Audiology,
18,
30,
(1958).
Effects of
in
(1979).
the
Download