Application of Noise Reduction Algorithm ClearVoice in Cochlear Implant Processing: Effects on Noise Tolerance and Speech Intelligibility in Noise in Relation to Spectral Resolution


Dingemanse, J.Gertjan.; Goedegebure, Aé.

Ear and Hearing 36(3): 357-367

2016


Noise reduction algorithms have recently been introduced in the design of clinically available cochlear implants. This study was intended to (1) evaluate the effect of noise reduction algorithm "ClearVoice" on noise tolerance and on speech intelligibility in noisy conditions at different speech-in-noise ratios in cochlear implant users, and (2) test the hypothesis that CI recipients with low spectral resolution might benefit more from noise reduction algorithms than CI users with high spectral resolution. A double-blind crossover design was used to measure the effect of the noise reduction algorithm ClearVoice on noise tolerance with the acceptable noise level (ANL) test and on speech in noise for three performance levels: speech reception thresholds (SRT) at 50%, 70%, and at a speech to noise ratio of SRT50% + 11 dB. Furthermore, they tested speech intelligibility in quiet. The effective spectral resolution was measured with a spectral-ripple discrimination test. Twenty users of the Advanced Bionics Harmony processor with HiRes120-processing participated in this study. The noise reduction algorithm led to a significant improvement-a decrease of 3.6 dB-in the ANL test but had no significant effect on any of the three speech-in-noise performance levels. The improvement in ANL was not significantly correlated with any of the speech-in-noise measures, nor with the speech-in-noise ratio in the ANL test. However, higher maximum speech intelligibility in quiet conditions correlated significantly with higher noise tolerance. Spectral-ripple discrimination thresholds were not significantly correlated with the effect of noise reduction on ANL or on speech intelligibility in noise nor with the speech-in-noise ratios. The spectral-ripple discrimination thresholds did correlate significantly with maximum speech intelligibility in quiet but not with speech reception thresholds in noise. The noise reduction algorithm ClearVoice improves noise tolerance. However, this study shows no change in speech intelligibility in noise due to the algorithm. The improvement in noise tolerance is not significantly related to spectral-ripple discrimination thresholds, speech intelligibility measures, or signal to noise ratio. Our hypothesis that CI recipients with low spectral resolution have a greater benefit from noise reduction than CI users with high spectral resolution does not hold for noise tolerance or for speech intelligibility in noise.

Application
of
Noise
Reduction
Algorithm
ClearVoice
in
Cochlear
Implant
Processing:
Effects
on
Noise
Tolerance
and
Speech
Intelligibility
in
Noise
in
Relation
to
Spectral
Resolution
J.
Gertjan
Dingemanse
and
Andre
Goedegebure
Objectives:
Noise
reduction
algorithms
have
recently
been
introduced
in
the
design
of
clinically
available
cochlear
implants.
This
study
was
intended
to
(1)
evaluate
the
effect
of
noise
reduction
algorithm
"ClearVoice"
on
noise
tolerance
and
on
speech
intelligibility
in
noisy
conditions
at
different
speech-in-noise
ratios
in
cochlear
implant
users,
and
(2)
test
the
hypoth-
esis
that
CI
recipients
with
low
spectral
resolution
might
benefit
more
from
noise
reduction
algorithms
than
CI
users
with
high
spectral
resolution.
Design:
A
double-blind
crossover
design
was
used
to
measure
the
effect
of
the
noise
reduction
algorithm
ClearVoice
on
noise
tolerance
with
the
acceptable
noise
level
(ANL)
test
and
on
speech
in
noise
for
three
perfor-
mance
levels:
speech
reception
thresholds
(SRT)
at
50%,
70%,
and
at
a
speech
to
noise
ratio
of
SRT50%
+
11
dB.
Furthermore,
they
tested
speech
intelligibility
in
quiet.
The
effective
spectral
resolution
was
measured
with
a
spectral-ripple
discrimination
test.
Twenty
users
of
the
Advanced
Bionics
Harmony
processor
with
HiRes120-processing
participated
in
this
study.
Results:
The
noise
reduction
algorithm
led
to
a
significant
improve-
ment—a
decrease
of
3.6
dB—in
the
ANL
test
but
had
no
significant
effect
on
any
of
the
three
speech-in-noise
performance
levels.
The
improvement
in
ANL
was
not
significantly
correlated
with
any
of
the
speech-in-noise
measures,
nor
with
the
speech-in-noise
ratio
in
the
ANL
test.
However,
higher
maximum
speech
intelligibility
in
quiet
conditions
correlated
significantly
with
higher
noise
tolerance.
Spectral-ripple
dis-
crimination
thresholds
were
not
significantly
correlated
with
the
effect
of
noise
reduction
on
ANL
or
on
speech
intelligibility
in
noise
nor
with
the
speech-in-noise
ratios.
The
spectral-ripple
discrimination
thresholds
did
correlate
significantly
with
maximum
speech
intelligibility
in
quiet
but
not
with
speech
reception
thresholds
in
noise.
Conclusions:
The
noise
reduction
algorithm
ClearVoice
improves
noise
tolerance.
However,
this
study
shows
no
change
in
speech
intelligibility
in
noise
due
to
the
algorithm.
The
improvement
in
noise
tolerance
is
not
significantly
related
to
spectral-ripple
discrimination
thresholds,
speech
intelligibility
measures,
or
signal
to
noise
ratio.
Our
hypothesis
that
CI
recipients
with
low
spectral
resolution
have
a
greater
benefit
from
noise
reduction
than
CI
users
with
high
spectral
resolution
does
not
hold
for
noise
tolerance
or
for
speech
intelligibility
in
noise.
Key
words:
Acceptable
noise
level,
ClearVoice,
Cochlear
implant,
Noise
reduction
algorithm,
Spectral-ripple
discrimination,
Speech
reception
threshold.
(Ear
&
Hearing
2015;36:357-367)
INTRODUCTION
With
current
cochlear
implants
(CIs),
recipients
can
under-
stand
speech
substantially
well
in
quiet,
but
when
there
is
back-
ground
noise
this
remains
difficult.
It
is
well
known
that
this
difficulty
is
a
major
complaint
of
most
hearing-impaired
people.
Department
of
ENT,
Erasmus
Medical
Center,
Rotterdam,
The
Netherlands.
For
this
reason,
much
of
the
current
hearing
device
research
aims
to
develop
technologies
that
improve
speech
intelligibility
in
noise
or
at
least
provide
more
listening
comfort.
Among
these
technolo-
gies
are
noise
reduction
algorithms
(NRAs).
These
algorithms
are
based
either
on
input
from
a
single-microphone
input
or
from
two
or
more
microphones.
Single-microphone
algorithms
perform
best
in
situations
with
stationary
noise,
whereas
multi-microphone
algorithms
work
best
in
conditions
where
speech
and
noise
come
from
different
directions
in
low-reverberant
surroundings
(e.g.,
Spriet
et
al.
2007;
Chung
et
al.
2012;
Hersbach
et
al.
2012;
Kokki-
nakis
et
al.
2012).
The
application
of
NRAs
in
commercially
avail-
able
CIs
is
only
a
recent
development.
ClearVoice,
a
proprietary
NRA
developed
by
Advanced
Bionics
(Valencia,
CA),
was
one
of
the
first
single-microphone
algorithms
applied
in
CIs.
According
to
evidence-based
medicine
principles,
when
using
NRAs
in
CI
sound
processing
it
is
important
to
gather
evidence
on
the
effects
of
these
algorithms
on
listening
comfort
and
speech
intelligibility
in
background
noise.
Moreover,
it
is
desirable
to
know
the
relevant
individual
factors
that
contribute
to
the
efficacy
of
NRAs.
ClearVoice
tries
to
distinguish
speech
and
noise
on
the
basis
of
different
temporal
and
spectral
characteristics.
Like
many
single-
microphone
algorithms,
ClearVoice
consists
of
three
elements:
(1)
an
estimation
of
the
noise
in
each
frequency
channel;
(2)
an
estimation
of
the
instantaneous
signal
to
noise
ratio
(SNR),
based
on
the
noise-estimate;
and
(3)
a
gain
calculation
for
the
attenu-
ation
of
spectral
channels
with
low
signal
to
noise
ratio.
Clear-
Voice
estimates
the
background
noise
level
with
a
minimum
tracking
method
in
a
time
window
of
1.3
sec
for
each
channel,
compares
the
actual
overall
level
within
a
channel
with
the
esti-
mated
noise
level,
and
attenuates
the
channel
if
its
actual
level
is
close
to
what
was
estimated.
ClearVoice
has
three
options
for
the
amount
of
attenuation
in
a
channel:
low
(up
to
—6
dB
attenuation),
medium
(up
to
—12
dB),
and
high
(up
to
—18
dB).
The
attenuation
is
applied
directly
to
the
electric
outputs.
This
avoids
limitations
in
the
reconstruction
of
an
acoustic
waveform
and
it
is
computation-
ally
efficient
(Advanced
Bionics
2012b;
Buechner
et
al.
2010).
For
CI
recipients,
several
studies
reported
that
single-micro-
phone
noise
reduction
techniques
improve
speech
intelligibility
in
background
noise
with
limited
temporal
fluctuations
(Hoch-
berg
et
al.
1992;
Toledo
et
al.
2003;
Yang
&
Fu
2005;
Kasturi
&
Loizou
2007;
Buechner
et
al.
2010;
Dawson
et
al.
2011;
Mauger
et
al.
2012).
The
reported
improvements
are
modest
to
small.
Others
found
no
effect
in
any
of
their
experiments
or
only
in
some
specific
conditions
(Hu
et
al.
2007;
Chung
et
al.
2012;
Kam
et
al.
2012;
Holden
et
al.
2013).
For
the
NRA
ClearVoice,
mixed
results
were
reported.
Buechner
et
al.
(2010)
found
a
significant
mean
improvement
of
20
percentage
points
for
0196/0202/2015/363-0357/0
Ear
&
Hearing
Copyright
©
2014
Wolters
Kluwer
Health,
Inc.
All
rights
reserved
Printed
in
the
U.S.A.
357
358
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
intelligibility
in
noise
with
ClearVoice
medium
and
24
percent-
age
points
with
ClearVoice
high.
He
tested
intelligibility
at
indi-
vidually
set
speech-to-noise
ratios
(within
the
0-6
dB
range)
in
a
sentence
test
with
the
level
of
stationary
speech-shaped
noise
set
at
55
dB.
Kam
et
al.
(2012)
found
a
small,
just
signifi-
cant
improvement
of
5.5
percentage
points
for
the
ClearVoice
Medium
setting
and
no
significant
effect
for
the
ClearVoice
High
setting
in
a
Cantonese
Hearing
in
noise
test
at
individually
set
speech-to-noise
ratios
(ranging
from
1
to
14.5
dB)
with
a
stationary
speech-shaped
noise
level
of
70
dB.
Advanced
Bion-
ics
investigated
the
benefits
of
ClearVoice
in
a
multicenter
study
(Advanced
Bionics
2012a).
After
a
2
week
period,
ClearVoice
medium
and
ClearVoice
high
were
evaluated
with
a
sentence-in-
noise
test
at
individually
set
speech-to-noise
ratios.
The
speech
level
was
set
at
60
dB
(SPL),
and
the
SNRs
were
in
the
range
of
2
to
10
dB
(reported
in
Holden
et
al.
2013).
Mean
percentage
correct
scores
improved
significantly—by
8.7
and
10.6
percent-
age
points,
respectively—with
ClearVoice
medium
and
Clear-
Voice
high
for sentences
in
stationary
speech-shaped
noise.
Holden et
al.
(2013)
evaluated
the
effect
of
ClearVoice
on
speech
recognition
in
multiple
noise
conditions,
including
res-
taurant
noise
(R-SPACETm),
stationary
speech-shaped
noise,
and
four-
and
eight-talker
babble.
Group
mean
scores
with
ClearVoice
medium
or
ClearVoice
high
were
not
significantly
different
from
the
control
condition,
except
for
ClearVoice
high
in
R-SPACE
noise.
For
this
condition,
a
2.5
dB
improvement
of
speech
reception
threshold
(SRT)
was
reported.
In
the
sen-
tence
test
with
stationary
speech-spectrum
noise,
the
speech
was
presented
at
50
dB
(SPL)
and
at
SNRs
in
the
range
of
2
to
8
dB.
The
effect
of
ClearVoice
in
pediatric
users
was
investi-
gated
by
Noel-Petroff
et
al.
(2013)
and
Schramm
et
al.
(2011).
Noel-Petroff
et
al.
reported
better
speech
intelligibility
in
con-
tinuous
speech-shaped
noise
on
a
sentence-in-noise
test
for
ClearVoice
High
after
a
1
month
period
of
ClearVoice
usage.
In
a
test
immediately
after
activation
of
ClearVoice,
no
significant
effect
was
seen.
Schramm
et
al.
reported
a
group
mean
improve-
ment
of
19.5
percentage
points
for
ClearVoice.
In
both
studies,
for
most
children,
the
T
and
M
levels
were
raised
according
to
feedback
from
the
child
so
as
to
maintain
the
most
comfortable
level.
Order
and
learning
effects
were
unable
to
be
ruled
out
in
either
study.
In
summary,
in
the
majority
of
the
ClearVoice
studies,
a
sig-
nificant
effect
of
ClearVoice
on
speech
intelligibility
in
noise
was
found.
Most
studies
reported
individual
outcomes
and
showed
large
variation
between
subjects.
Furthermore,
the
stud-
ies
differed
in
a
number
of
aspects,
which
makes
their
results
difficult
to
compare.
These
differences
included
sound
level;
speech
and
noise
material;
study
design
aspects
like
power
anal-
ysis,
blinding,
and
test
order;
allowance
of
changes
in
volume
settings;
and
changes
in
M
or
T
levels.
In
this
study,
we
wanted
to
test
ClearVoice
in
a
well-designed
experiment,
with
the
main
focus
being
on
the
effect
of
ClearVoice
on
noise
tolerance.
Fur-
thermore,
we
wanted
to
search
for
an
explanation
for
the
large
differences
between
subjects
in
the
effect
of
ClearVoice.
It
is
not
clear
yet
why
NRAs
in
CIs
improve
speech-in-noise
scores
and
why
there
are
large
differences
between
individu-
als.
Hu
et
al.
(2007)
believed
that
much
of
the
success
of
the
NRA
in
CI
processing
can
be
attributed
to
the
improved
tempo-
ral
envelope
contrast.
Chung
et
al.
(2006)
hypothesized
that
for
CI
users
the
improvement
comes
from
the
fact
that
the
band-
width
of
the
frequency
channels
in
CI
processing
is
narrower
than
the
effective
bandwidth
of
the
CI
stimulation.
If
one
of
the
frequency
channels
is
noise
dominated
and
attenuated
but
the
neighboring
band
is
not,
then
the
effective
signal
to
noise
ratio
in
the
broader
frequency
band
of
CI
stimulation
is
improved
when
both
processing
channels
fall
into
the
same
stimulation
band.
Based
on
the
explanation
of
Chung
and
colleagues,
we
hypothesized
that
CI
recipients
with
low
spectral
resolution
might
have
more
benefit
from
noise
reduction
than
CI
users
with
high
spectral
resolution.
This
hypothesis
could
explain
the
large
intersubject
variability
in
the
effect
of
NRAs.
To
test
this
hypothesis,
we
decided
to
use
a
spectral-ripple
(SR)
dis-
crimination
test
as
a
measure
of
spectral
resolution.
The
SR
dis-
crimination
test
evaluates
the
ability
of
a
listener
to
discriminate
between
standard
and
inverted
rippled
spectra,
and
the
outcome
measure
is
the
minimum
ripple
spacing
discerned
by
listeners.
In
the
recent
literature,
there
has
been
a
debate
about
unwanted
cues
such
as
local
loudness
cues,
spectral
boundary
cues,
and
spectral
centroid
cues
that
CI
recipients
might
use
in
a
SR
dis-
crimination
test
(Azadpour
and
McKay
2012;
Jones
et
al.
2013).
Several
studies
confirmed
that
the
SR
measurement
is
related
to
spectral
resolution
when
using
current
clinical
CIs
(Anderson
et
al.
2011;
Won
et
al.
2011;
Jones
et
al.
2013).
The
minimum
dis-
cerned
SR
spacing
correlates
with
vowel
and
consonant
recog-
nition
in
quiet
(Henry
and
Turner
2003;
Henry
et
al.
2005)
and
with
word
recognition
in
quiet
and
in
noise
(Won
et
al.
2007).
In
contrast,
Anderson
et
al.
(2011)
found
no
correlation
between
SR
discrimination
thresholds
and
SRTs
for
words
in
sentences
or
for
vowel
recognition
in
noise.
In
the
quiet
condition,
they
found
that
words
in
sentences
and
SR
discrimination
thresholds
were
significantly
correlated.
Besides
speech
enhancement
in
noise,
another
important
effect
of
NRAs
is
that
they
improve
aspects
of
listening
com-
fort,
such
as
noise
tolerance
and
ease
of
listening
(Ricketts
&
Hornsby
2005;
Bentler
et
al.
2008;
Zakis
et
al.
2009;
Luts
et
al.
2010).
Ricketts
and
Hornsby
(2005)
and
Luts
et
al.
(2010)
used
paired
comparisons
and
found
a
preference
for
noise
reduction
over
the
unprocessed
condition
for
most
NRAs
among
both
impaired
and
normal-hearing
listeners.
Bentler
et
al.
(2008)
documented
significantly
better
ease
of
listening
ratings
among
hearing-impaired
listeners
for
conditions
with
noise
reduction.
Luts
et
al.
reported
a
reduction
of
perceived
listening
effort
at
0
dB
SNR
for
noise
reduction
in
comparison
with
a
control
condition.
For
CI
users,
the
effects
of
NRAs
on
noise
tolerance
and
listening
effort
are
not
well
documented.
Only
sound
quality
preferences
(Chung
et
al.
2006,
2012)
or
preferences
for
noise
reduction
in
daily
life
were
reported.
The
percentage
of
partici-
pants
that
reported
a
preference
for
a
ClearVoice
program
was
53%
in
the
Buechner
et
al.
(2010)
study
and
66%
in
the
Kam
et
al.
(2012)
study.
Furthermore,
Buechner
et
al.
collected
subjec-
tive
ratings
of
the
programs
in
every
day
listening
situations
with
the
Abbreviated
Profile
of
Hearing
Aid
Benefit
(APHAB)
and
found
no
significant
difference
in
scores
between
programs
with
ClearVoice
on
and
those
in
which
it
was
off.
To
evaluate
the
increase
in
noise tolerance
due
to
NRAs,
the
acceptable
noise
level
(ANL)
test
is
often
used.
In
1991,
Nabelek
and
colleagues
developed
this
procedure
for
the
determination
of
ANLs
while
listening
to
speech
(Nabelek
et
al.
1991).
The
ANL
procedure
quantifies
a
listener's
willingness
to
listen
to
speech
in
the
presence
of
background
noise.
To
obtain
an
ANL
measurement,
a
recorded
story
of
running
speech
is
adjusted
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,357-367
359
to
the
listener's
most
comfortable
listening
level
(MCL).
Next,
background
noise
is
added
and
adjusted
to
a
level
(called
back-
ground
noise
level
[BNL])
that
the
listener
is
willing
to
"put
up
with"
while
listening
to
and
following
the
words
of
the
story.
The
ANL
is
calculated
by
subtracting
the
BNL
from
the
MCL
and
is
the
lowest
SNR
that
a
listener
is
willing
to
accept.
Low
ANL
val-
ues
indicate
a
high
tolerance
of
background
noise,
whereas
high
values
indicate
a
low
tolerance.
ANL
is
not
related
to
a
speech
intelligibility
in
noise
task,
and
the
difference
between
unaided
and
aided
ANL
is
very
small
(Nabelek
et
al.
2004,
2006).
The
effect
of
NRAs
on
ANL
in
hearing
aids
users
has
been
evaluated
in
a
few
studies
(Mueller
et
al.
2006;
Peeters
et
al.
2009;
Pisa
et
al.
2010).
Mueller
et
al.
showed
a
mean
improvement
of
4.2
dB
for
the
ANL,
Peeters
et
al.
observed
a
mean
improvement
of
3.3
dB,
and
Pisa
et
al.
reported
a
mean
improvement
of
1.2
dB.
These
studies
used
steady
state
speech-spectrum
noise.
Holden
et
al.
(2013)
administered
the
ANL
test
to
CI
users
with
Clear-
Voice
off,
ClearVoice
medium,
and
ClearVoice
high.
They
used
running
speech
in
a
12-talker
babble.
They
did
not
find
signifi-
cant
group
differences
between
the
conditions.
The
main
research
question
of
this
study
was
as
follows:
What
is
the
effect
of
the
clinically
available
single-microphone
NRA
ClearVoice
on
noise
tolerance
and
on
speech
intelligi-
bility
in
noise
among
CI
users?
The
noise
tolerance
was
mea-
sured
with
an
ANL
test,
and
SRTs
were
adaptively
estimated
at
percent
correct
levels
of
50%
and
70%,
called
SRT50%
and
SRT70%.
Furthermore,
a
speech
intelligibility
level
(percent
correct)
was
measured
at
an
SNR
of
11
dB
above
the
SRT50%.
We
also
included
a
measurement
of
speech
intelligibility
in
quiet.
We
added
a
questionnaire
about
perceived
problems
in
daily
life
communication
for
correlation
with
the
ANL
and
speech-in-noise
scores.
The
secondary
question
was
whether
the
intersubject
vari-
ability
in
the
effect
of
the
NRA
on
ANL
and
speech
recognition
in
noise
might
be
related
to
the
spectral
resolution
of
the
CI
stimulation
as
measured
with
a
SR
discrimination
test.
MATERIALS
AND
METHODS
Study
Design
The
NRA
that
was
investigated
in
this
study
is
Clear-
Voice.
It
is
a
proprietary
single-microphone
NRA
developed
by
Advanced
Bionics
LLC,
which
works
together
with
their
HiRes
Fidelity
120
technology.
The
details
of
the
algorithm
are
described
in
the
Introduction.
In
this
study,
we
used
the
medium
setting
of
ClearVoice.
All
participants
were
tested
with
the
same
new Harmony
processor
and
a
new
T-mic
(Advanced
Bionics).
For
several
reasons,
no
adjustments
of
M
and
T
levels
or
volume
settings
were
made
during
testing.
First,
from
a
scientific
point
of
view,
we
preferred
to
test
the
effect
of
the
NRA
alone,
instead
of
the
combined
effect
of
the
NRA
and
level
adjustments.
Sec-
ond,
in
practice,
many
CI
users
do
not
switch
their
program
or
change
their
volume
setting
depending
on
the
noise
situation
or
even
when
they
change
from
noisy
to
quiet
surroundings.
Therefore,
we
felt
that
a
standard
increase
in
M
level
was
not
appropriate.
The
effect
of
noise
reduction
on
ANL,
SRT,
and
percent
cor-
rect
(Pc)
words
was
investigated
in
a
crossover
design.
Because
this
type
of
design
has
a
risk
of
introducing
order
effects,
like
a
learning
effect
or
a
fatigue
effect,
we
included
an
evaluation
of
order
effects
in
the
statistical
analyses
of
the
results.
The
different
tests
were
allocated
into
two
separate
test
ses-
sions.
The
second
session
was
2
to
7
days
after
the
first.
The
first
session
consisted
of
three
blocks.
In
the
first
block,
we
measured
the
SRT
for
50%
correct
to
make
the
participants
familiar
with
the
task
and
to
obtain
a
first
estimation
of
a
participants
SRT50%,
which
we
called
SRT50%leam.
Second,
we
measured
the
maxi-
mum
percent
correct
score
at
a
signal
to
noise
ratio
of
40
dB.
In
the
second
and
third
blocks,
the
effect
of
the
NRA
was
tested
with
three
speech-in-noise
conditions,
including
SRTs
at
target
scores
of
50%
and
70%
and
percent
correct
scores
at
a
fixed
speech-in-
noise
ratio
of
SRT50%learn
+
11
dB.
Within
a
block,
the
three
conditions
were
tested
in
a
randomly
interleaved
order.
At
the
end
of
the
second
block,
we
again
measured
the
maximum
percent
correct
score
at
a
signal
to
noise
ratio
of
40
dB.
All
participants
used
a
CI
with
three
user
programs.
We
asked
each
participant
which
of
the
programs
he
or
she
used
most
often
in
everyday
life
situations.
The
settings
of
this
pro-
gram
were
placed
into
each
of
the
three
programs
with
Clear-
Voice
off
(condition
NRA-off).
Then,
a
clinician
other
than
the
test
examiner
switched
ClearVoice
on
(condition
NRA-on)
in
either
program
2
or
program
3
for
comparison
between
NRA-
off
and
NRA-on.
The
clinician
did
this
in
a
quasi-random
order,
so
that
in
the
end
10
participants
had
the
noise
reduction
on
in
program
2
and
10
in
program
3.
During
the
experiment,
the
participants
used
program
1
in
test
block
1
(NRA-off),
program
2
in
test
block
2,
and
program
3
in
test
block
3.
This
procedure
was
intended
to
create
a
double-blind
situation.
"Blinding"
of
participants
means
that
they
were
not
informed
about
the
noise
reduction
setting.
However,
we
were
unable
to
rule
out
that
the
attenuation
of
the
noise
by
the
NRA
may
have
been
audible.
To
minimize
this
potential
influence,
interleaved
testing
of
differ-
ent
SNR
conditions
was
applied,
to
make
the
detection
of
the
noise
reduction
condition
more
difficult.
The
second
test
session
consisted
of
the
ANL
test
and
a
SR
test.
The
details
of
the
tests
are
described
later.
First,
a
practice
condition
of
the
ANL
test
was
done
with
CI
program
1,
fol-
lowed
by
two
practice
runs
of
the
SR
test.
Then,
the
ANL
test
was
performed
with
CI
program
2
and
CI
program
3
for
com-
parison
between
NRA-off
and
NRA-on
conditions.
After
the
ANL
test
and
a
pause,
three
runs
of
the
SR
test
were
performed
with
CI
program
1.
Participants
Twenty
users
of
an
Advanced
Bionics
CI
system
(HiRes
90K
implant
and
Harmony
processor)
participated
in
this
study.
The
ages
of
the
participants
ranged
from
37
to
85
years,
with
a
mean
of
65
years.
All
participants
had
used
16
active
electrodes
and
HiRes120
sound
processing
for
at
least
1
year.
All
participants
are
unilateral
CI
users
with
a
group
mean
of
4.2
(SD
2.0)
years
of
CI
use.
All
but
two
used
the
NRA
ClearVoice
in
their
daily
program.
The
input
dynamic
range
(IDR)
setting
was
between
55
and
65
dB
(2
x
55
dB;
15
x
60
dB;
3
x
65
dB).
Some
partici-
pants
wear
a
hearing
aid
in
the
nonimplanted
ear,
but
they
did
not
wear
it
during
the
tests.
All
participants
were
Dutch
native
speakers
who
reported
normal
reading
ability.
For
inclusion
in
this
study,
a
phoneme
score
of
at
least
80%
on
clinically
used
Dutch
consonant-vowel-consonant
word
lists
was
required.
Participants
were
required
to
sign
a
written
informed
consent
form
before
participating
in
the
study.
Approval
of
the
Erasmus
Medical
Center
Ethics
Committee
was
obtained.
360
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
Equipment
The
test
was
set
up
in
a
sound-treated
room
in
the
department
of
ENT/Audiology
of
the
Erasmus
Medical
Center.
Test
partici-
pants
sat
1
m
in
front
of
a
loudspeaker
that
was
connected
to
a
Madsen
OB822
audiometer,
a
Behringer
UCA202
soundcard,
and
a
Macbook
pro
(type
A1278)
notebook.
Data
interpretation
and
analysis
were
done
with
Matlab
(v7.11.0)
and
SPSS
(v20).
ANL
Test
The
ANL
is
the
difference
between
the
MCL
for
running
speech
and
a
BNL.
The
ANL
was
tested
with
the
same
speech
and
noise
material
as
the
speech
intelligibility
in
noise
test.
The
sentences
were
connected
with
intervals
of
500
msec
of
silence
between
them
and
played
as
running
speech.
The
listeners
were
given
written
instructions,
which
were
Dutch
translations
of
the
instructions
in
Nabelek
et
al.
(2006),
and
MCL
and
BNL
were
obtained
according
to
previous
ANL
research
(e.g.,
Nabelek
et
al.
2004).
In
a
practice
condition,
MCL
and
BNL
were
determined
twice.
In
the
test
conditions,
the
MCL
and
BNL
procedures
were
repeated
three
times,
and
the
mean
values
were
used
for
calculation
of
the
ANL
and
for
data
analysis.
Speech-in-Noise
Test
Speech
reception
was
measured
with
Dutch
female-spoken,
unrelated
sentences
of
five
to
nine
words
(with
a
median
length
of
six
words)
in
steady
state
speech-spectrum
noise
(Versfeld
et
al.
2000).
For
each
condition,
two
lists
of
13
sentences
were
used.
The
presentation
level
of
the
sentences
was
fixed
at
70
dB
(SPL).
The
noise
started
3
sec
before
the
speech
and
ended
0.5
sec
after
the
speech.
Participants
were
asked
to
repeat
as
many
words
of
the
sentence
they
understood,
after
a
brief
tone
that
was
given
3
sec
after
the
end
of
each
sentence.
A
percentage
of
correct
words
per
sentence
list
was
calculated.
Speech
perception
in
noise
was
mea-
sured
at
three
SNRs
and
three
different
performance
levels.
The
SRT
for
50%
(SRT50)
and
70%
(SRT70)
were
measured
with
an
adaptive
procedure.
In
addition,
the
percentage
correct
was
measured
at
a
fixed
signal
to
noise
ratio
of
SRT50%learn
+
11
dB.
The
adaptive
procedure
we
used
was
a
stochastic
approxima-
tion
method
with
step
size
4
(Pc(n
1)
target_Pc)
(Robbins
&
Monro
1951),
with
Pc(n
1)
being
the
percent
correct
score
of
the
previous
trial.
The
SRT
was
defined
as
the
average
SNR
over
the
last
23
presentation
levels.
(The
27th
level
was
calcu-
lated
from
the
response
on
the
26th
sentence.)
It
was
proven
that
the
average
of
trials
in
a
stochastic
approximation
staircase
with
constant
step
size
converges
to
the
target
(Kushner
&
Yin
2003).
The
maximum
percentage
correct
was
measured
at
an
SNR
of
40
dB.
This
is
equivalent
to
the
measurement
of
percentage
cor-
rect
in
quiet,
but
it
has
the
advantage
that
it
is
a
distinct
point
on
the
psychometrical
curve,
instead
of
being
the
asymptotic
value.
SR
Test
For
the
SR
test,
noise
stimuli
were
generated
which
had
log-
arithmically
spaced
SRs
using
the
following
equation:
D
—sint2fr-log
2
(f/L)-fs+0
0
}/20
X
(f)=
10
2
where
X(f)
is
the
amplitude
of
a
bin
with
center
frequencyfHz,
D
is
the
spectral
modulation
depth
or
peak-to-valley
ratio
(in
dB),
L
is
the
low
cutoff
frequency
of
the
noise
passband,
fs
is
the
spectral
modulation
frequency
in
ripples/octave,
and
e
o
is
the
starting
phase
of
the
spectral
modulation
(Litvak
et
al.
2007;
Anderson
et
al.
2011).
Next,
the
magnitudes
of
the
frequency
components
were
shaped
according
to
the
long-term
average
speech-spectrum
of
the
sentences
used
in
the
speech-in-noise
test.
The
low
cutoff
frequency
L
was
100
Hz,
and
the
high
cutoff
frequency
was
8000
Hz.
The
spectral
modulation
depth
D
was
30
dB
except
for
the
edges
of
the
SR,
where
cosine-shaped
ramps
with
a
length
of
1/3
octave
were
applied
to
prevent
for
unwanted
cues
at
the
frequency
boundaries.
Stimuli
were
generated
in
the
frequency
domain
assuming
a
sampling
rate
of
44,100
Hz
and
a
signal
duration
of
500
msec.
The
starting
phases
of
the
individ-
ual
frequency
components
were
randomized
for
each
stimulus
and
trial
to
avoid
fine
structure
pitch
cues
that
might
have
been
perceptible
to
listeners.
The
starting
phase
of
the
spectral
modu-
lation
e
o
was
selected
at
random,
with
a
uniform
distribution
(0
to
27c
rad)
for
each
trial.
This
randomization
was
intended
to
limit
the
ability
of
listeners
to
rely
exclusively
on
a
certain
fre-
quency
channel
to
perform
SR
discrimination
at
a
certain
ripple
density.
For
inversely
rippled
noise,
the
starting
phase
for
the
spectral
modulation
was
e
o +
at.
After
taking
an
inverse
Fourier
transform,
100
msec
cosine-shaped
onset
and
offset
ramps
were
applied.
The
sentences
were
filtered
to
a
100
to
8000
Hz
pass-
band,
and
after
the
filtering,
the
long-term
RMS
value
of
the
amplitude
was
obtained.
The
SR
stimuli
were
given
the
same
RMS
value
and
played
with
the
same
calibration
and
signal
path
as
the
speech
at
70
dB
(SPL).
To
reduce
cues
related
to
loud-
ness,
the
noise
level
was
roved
across
intervals
within
each
trial
by
—3
dB
or
+3
dB.
The
design
of
the
SR
stimuli
prevents
the
detection
of
spectral
boundary
cues,
loudness
cues,
or
spectral
centroid
cues
and
is
in
accordance
with
the
stimuli
of
Won
et
al.
(2011)
and
Jones
et
al.
(2013).
They
validated
that
the
SR
test
with
these
stimuli
is
related
to
spectral
resolution
when
used
with
the
Advanced
Bionics
HiRes90K
implant.
Spectral
modulation
thresholds
were
determined
using
a
cued
adaptive
three-interval,
two-alternative
forced
choice
(3I-2AFC)
procedure.
The
interstimulus
interval
was
500
msec.
The
inverted
ripple
was
randomly
presented
in
one
of
the
inter-
vals.
The
subject
was
asked
to
choose
the
interval
that
sounded
different.
A
one-up,
three-down
stepping
rule
was
used
with
an
increasing
and
decreasing
factor
of
1.41.
With
this
stepping
rule,
the
masked
threshold
at
79.4% correct
discrimination
was
estimated.
Each
test
run
started
at
a
ripple
rate
of
0.25
ripples
per
octave.
The
run
was
terminated
after
10
reversals,
and
the
geometric
mean
ripple
rate
at
the
last
six
reversal
points
was
used
to
determine
the
threshold
for
ripple
discrimination.
Two
practice
runs
and
three
test
runs
were
performed.
The
mean
SR
threshold
was
calculated
as
the
geometric
mean
ripple
rate
of
the
three
test
runs.
APHAB
Questionnaire
All
participants
were
asked
to
complete
the
APHAB,
a
24-item
questionnaire
to
assess
the
participants'
experience
with
CI
use
in
everyday
communication
situations
(Cox
&
Alexander
1995).
The
APHAB
has
a
Global
score
of
all
questions
and
four
subscales:
ease
of
communication,
speech
recognition
in
rever-
beration
and
in
background
noise,
and
aversiveness
of
sound.
Participants
answered
for
each
of
the
24
items
how
often
a
state-
ment
was
true
in
daily
communication
by
making
a
choice
between
I
I
NRA
off
NRA
on
3.6
dB
p
<
0.001
NRA
off
NRA
on
A
A
A
A
#
A
la
A
-0
z
20
15
10
5
0
-5
wor
d
sco
re
(
%)
100
80
60
40
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
361
the
options
always
(approximately
99%
of
the
time),
generally
(75%),
half
of
the
time
(50%),
occasionally
(25%),
or
never
(1%).
RESULTS
Acceptable
Noise
Level
The
group
mean
value
of
the
MCL
was
61.1
dB
(SD
5.6)
for
the
NRA-on
and
61.2
dB
(SD
5.6)
for
the
NRA-off
condition.
The
difference
between
the
NRA-on
and
NRA-off
condition
was
not
significant
(difference
=
0.09,
p
=
0.7),
indicating
that
the
NRA
had
no
effect
on
perceived
loudness
for
speech
signals.
Figure
1
shows
the
group
mean
ANL
values
for
both
condi-
tions.
A
normality
check
revealed
that
the
ANL
data
could
be
regarded
as
having
a
normal
distribution.
With
NRA-on
par-
ticipants
accepted
more
noise
than
in
the
NRA-off
condition.
A
paired
t
test
showed
that
the
ANL
value
for
the
NRA-on
condi-
tion
was
significantly
lowered
by
3.6
dB
(p
<
0.001).
We
determined
whether
an
order
effect
was
present.
An
analy-
sis
of
variance
with
a
between-subject
factor
"order"
and
a
within-
subject
factor
NRA
showed
neither
a
significant
effect
of
the
order
factor
[F(1,17)
=
1.2,
MSE
=
52.0,
p
=
0.30]
nor
an
effect
of
the
NRA
x
order
interaction
[F(1,17)
=
1.9,
MSE
=13.6,
p
<
0.19].
Figure
2
shows
the
individual
ANL
values
for
the
NRA-off
and
NRA-on
conditions
and
the
calculated
speech
intelligibility
at
the
mean
ANL
value,
this
being
the
mean
of
the
NRA-off
and
NRA-on
conditions.
This
word
score
was
calculated
from
indi-
vidual
logistic
functions
fitted
to
the
speech
intelligibility
data.
All
participants
had
word
scores
above
50%
at
their
ANLmean
value,
except
participant
3.
This
participant
apparently
used
a
different
criterion,
namely
how
much
noise
he
was
willing
to
accept,
without
listening
to
the
speech.
Therefore,
we
decided
to
exclude
the
ANL
data
of
participant
3.
The
ANL
difference
of
participant
2
deviated
by
2.9
SD
from
the
mean
ANL
difference.
If
we
exclude
participant
2,
the
mean
ANL
difference
due
to
the
NRA
is
4.2
dB.
We
questioned
whether
the
amount
of
ANL
improvement
due
to
noise
reduction
might
be
related
to
the
signal
to
noise
18
17
16
15
14
z
13
12
11
10
9
ANL
Fig.
1.
Mean
acceptable
noise
level
(ANL)
values
for
the
noise
reduction
algorithm
(NRA)
conditions
NRA-off
and
NRA-on.
Error
bars
represent
95%
confidence
intervals.
ratio
in
the
test.
However,
correlation
analyses
showed
no
sig-
nificant
relationship
between
the
ANL
difference
(ANLdiff)
and
the
mean
ANL
value
(ANLmean)
for
the
NRA-on
and
NRA-off
conditions
(Spearman
p
=
—0.03,
p
>
0.75).
Also,
the
correlation
coefficients
between
the
ANL
measures
(ANLdiff,
ANLmean,
MCL)
and
speech
intelligibility
in
noise
measures
were
calculated.
Results
of
the
calculation
did
not
show
significant
correlations
except
for
the
correlation
between
ANLmean
and
the
rationalized
arcsine
units
(rau)
scores
for
percentage
of
correct
words
at
an
SNR
of
40
dB
(Spearman
p
=
—0.52,
p
<
0.02).
Lower
rau
word
scores
in
(nearly)
quiet
situations
were
associated
with
higher
ANLmean
values.
Speech
Intelligibility
in
Noise
A
normality
check
of
the
SRT
data
for
50%
and
70%
correct
revealed
normally
distributed
data,
except
for
SRT70%
in
the
NRA-off
condition.
This
was
due
to
an
outlier
for
participant
13.
His
SRT70%
value
deviated
more
than
3
SDs
from
the
mean
3
9
6
1
5
8
19
12
20
15
17
2
10
7
14
11
16
4
18
13
participant
number
I I
I
I
I I I I I
•I
-
411
word
score
at
the
SNR
of
the
mean
ANL
over
'NRA
off
and
'NRA
on'
conditions
3
9
6
1
5
8
19
12
20
15
17
2
10
7
14
11
16
4
18
13
participant
number
Fig.
2.
Upper
panel:
individual
acceptable
noise
level
(ANL)
values
for
the
noise
reduction
algorithm
(NRA)
conditions
NRA-off
and
NRA-on.
Lower
panel:
Calculated
intelligibility
level
that
participants
used
in
the
ANL
test
(see
text
for
details
of
calculation).
wo
r
d
sco
re
(
rau
)
110
100
90
80
70
60
50
98.4
94.3
88.2
80,3
71
60.7
50
wor
d
sc
ore
(
%)
NRA
off;
SRT50%
NRA
on;
SRT50%
A
A
Alel
A
1
1
1
I
1
I
1
9
11
6
15 18
7
2
19
4
3
I I I
I
I
I
I
8
12
5
1
16
10
14 17
20
13
A
A
1
I
4
A
NRA
off;
SNR
=
40dB
-
362
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
NRA
off
A
NRA
on
-10
0
10
20
30
40
SNR
(dB)
Fig.
3.
Mean
results
of
speech
intelligibility
in
noise
tests
measured
for
noise
reduction
algorithm
(NRA)
conditions
NRA-off
(circles)
and
NRA-on
(tri-
angles)
with
95%
confidence
intervals.
The
points
for
50%
and
70%
correct
were
measured
with
an
adaptive
procedure
that
estimated
the
speech-to-
noise
ratio
(SNR).
The
other
points
were
measured
at
an
individualized
fixed
SNR.
The
percent
correct
scores
were
converted
to
rau
scores.
The
right
axis
shows
the
corresponding
word
scores
on
a
percent
correct
scale.
SRT70%
value.
We
excluded
participant
13
for
the
SRT70%
NRA-off
condition.
For
participant
20,
the
SRT70%
value
could
not
be
obtained
with
the
adaptive
test.
The
target
of
70%
was
too
close
to
his
maximum
percent
word
score
(73%).
The
percent
correct
data
were
transformed
to
rau
(Studebaker
1985).
Figure
3
shows
the
mean
values
and
95%
confidence
intervals
for
the
different
speech-in-noise
conditions
for
both
the
NRA-
on
and
NRA-off
conditions.
No
systematic
difference
between
the
NRA-on
and
the
NRA-off
data
points
was
observed.
For
the
three
conditions
we
measured
with
NRA-off
and
NRA-on,
we
performed
a
repeated
measures
analysis
of
variance
with
within-subject
factor
NRA
and
between-subject
factor
"order."
The
second
factor
was
added
to
investigate
if
learning
or
fatigue
effects
had
influenced
the
measurements.
No
significant
effect
on
the
NRA
factor
was
observed
in
any
of
the
conditions.
(SRT50%:
[F(1,18)
=
1.4,
MSE
=
1.2,
p
=
0.26];
SRT70%:
[F(1,16)
=
2.2,
MSE
=
3.3,
p
=
0.16];
Rau@SRT50%plldB:
[F(1,18)
=
0.55,
MSE
=
0.001,p
=
0.48]).
Also,
the
order
factor
was
not
significant
for
any
condition
(SRT50%:
[F(1,18)
=
1.6,
MSE
=
41.2,
p
=
0.22];
SRT70%:
[F(1,16)
=
2.2,
MSE
=
40.2,
p=
0.16];
Rau@SRT50%plldB:
[F(1,18)
=
1.2,
MSE
=
0.006,
p
=
0.30]).
The
curves
in
Figure
3
show
the
average
psychometric
curves
that
relate
word
scores
with
SNR.
We
fitted
a
logistic
function
to
the
individual
psychometric
curves
and
calculated
the
mean
slope.
The
mean
of
the
slope
around
the
50%
level
is
6.4%/dB
with
a
standard
deviation
of
2.1%/dB.
The
mean
of
the
maximum
percent
correct
scores
at
an
SNR
of
40
dB
was
94.3%.
Figure
4
shows
individual
data
points
for
SRT50%
for
NRA-
off
and
NRA-on
in
the
upper
axis.
The
range
of
SRT50%
is
approximately
from
0
to
15
dB,
but
the
majority
of
SRT50%
scores
was
between
1
and
5
dB.
SRTs
with
and
without
NR
were
highly
correlated
for
the
whole
SNR
range.
Furthermore,
Figure
4
shows
individual
percent
correct
scores
at
an
SNR
of
40
dB.
Comparison
of
both
panels
in
Figure
4
demonstrates
that
par-
ticipants
who
had
higher
SNR50%
tended
to
have
a
lower
word
score
at
an
SNR
of
40
dB.
The
Pearson
correlation
coefficient
for
SNR50%
and
rau-converted
word
scores
is
0.69
(p
<
0.001).
SR
Thresholds
A
secondary
purpose
of
this
study
was
to
test
the
hypoth-
esis
that
an
improvement
in
ANL
scores
or
speech
intelligibility
scores
due
to
noise
reduction
is
related
to
SR
thresholds.
Fig-
ure
5
shows
the
mean
of
the
log
2
transformed
values
of
the
SR
thresholds
for
each
participant
in
the
left
panel.
The
thresholds
were
log
2
transformed
to
make
them
normally
distributed.
For
participant
5,
we
had
only
one
SR
threshold
due
to
time
restric-
tions.
We
therefore
decided
to
exclude
this
participant
from
the
spectra-ripple
data
set.
The
SR
thresholds
varied
substan-
tially
between
participants,
which
allowed
us
to
investigate
the
15
participant
number
100
----
90
0
®
80
70
60
9
11
6
15 18
7
2
19
4
3
8
12
5
1
16
10
14 17
20
13
participant
number
Fig.
4.
Upper
panel:
individual
speech
reception
thresholds
for
the
performance
level
of
50%
correct
word
score
for
noise
reduction
algorithm
(NRA)
condi-
tions
NRA-off
and
NRA-on.
Lower
panel:
individual
maximum
word
scores
at
a
speech-to-noise
ratio
(SNR)
of
40
dB.
ANL
indicates
acceptable
noise
level.
wo
r
d
sco
re
a
t
S
NR
40dB
(
%)
...........
1
O
3
2019111415
5
8
2
131710
6
16
9
7
12
3
18
4
1
participant
number
R
2
=0.5
p
<
0.001
1
2
3
4
5
6
spectral
ripples
(ripples/octave)
0.95
0.9
0.85
0.8
0.75
0.7
0
0
Gb
0
0
0
0
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
363
Fig.
5.
Left
panel:
mean
of
the
log,
transformed
values
of
the
spectral-ripple
thresholds
in
ascending
order.
Error
bars
give
95%
confidence
intervals.
Right
panel:
relation
between
word
scores
at
a
speech-to-noise
ratio
(SNR)
of
40
dB
and
the
spectral-ripple
thresholds.
relation
between
the
SR
thresholds
and
the
performance.
The
mean
SR
threshold
was
1.8
ripples/octave.
We
analyzed
the
hypothesized
relation
between
the
SR
thresholds
and
the
effect
of
noise
reduction
on
ANL
and
speech
intelligibility.
No
significant
correlation
was
found
between
the
ANL
benefit
and
the
SR
thresholds.
Although
we
did
not
find
a
significant
mean
improvement
of
speech
intelligibility
due
to
noise
reduction,
we
calculated
the
correlation
between
SR
thresholds
and
the
difference
of
the
NRA-on
and
NRA-off
speech
measures.
The
correlation
was
insignificant
in
all
three
speech
performance
levels.
Because
we
expected
a
relation
between
spectral
resolu-
tion
and
general
performance,
we
correlated
the
SR
thresholds
with
the
different
speech-in-noise
outcome
measures
and
with
MCLmean
and
ANLmean.
Only
a
nearly
significant
correlation
was
found
between
the
SR
thresholds
and
the
rau
scores
for
percent
correct
words
at
an
SNR
of
40
dB
(Spearman
p
=
0.43,
p
<
0.07).
Better
spectral
resolution
was
related
to
higher
per-
centages
of
correct
scores
at
an
SNR
of
40
dB.
The
relationship
appeared
to
be
nonlinear.
We
applied
a
model
of
the
form:
Pc
=
1
-
a
/SR
This
simple
model
converges
to
100%
correct
for
high
SR
scores
and
to
0%
correct
for
very
low
SR
scores.
The
result
of
the
fit
is
a
value
of
0.7.
This
model
accounts
for
50%
of
the
variance
in
the
data
(R
2
=
0.5,
F[18]
=
7710,
MSE
=
0.0027,
p
<
3.73e-25)
and
confirmed
the
expected
relation
between
spectral
resolution
and
general
performance.
Between
the
SR
discrimination
threshold
and
SRT50%,
a
trend
was
seen
in
which
better
SR
thresholds
were
associated
with
better
speech-in-noise
thresholds,
but
the
correlation
was
not
significant
(Pearson
r
=
-0.30,p
=
0.21).
No
significant
cor-
relation
was
found
between
SR
thresholds
and
ANL.
Abbreviated
Profile
of
Hearing
Aid
Benefit
Table
1
shows
the
mean
APHAB
scores
and
the
Pearson
correlation
coefficients
examining
the
relationships
among
APHAB
scales,
speech
intelligibility
in
noise
(SRT50%),
and
ANLmean
scores.
Higher
scores
reflect
a
greater
frequency
of
perceived
problems
in
everyday
life
situations.
CI
users
perceived
most
frequent
communication
problems
in
reverberant
environments
and
in
situations
with
background
noise.
The
correlation
analysis
showed
that
higher
speech-
in-noise
thresholds
(SRT50%)
and
higher
ANL
values
were
TABLE
1.
Mean
scores
(with
standard
deviations)
on
the
Abbreviated
Profile
of
Hearing
Aid
Benefit
that
report
the
percentage
of
problems
on
subscales
ease
of
communication,
background
noise, reverberation,
aversiveness,
and
the
global
scale
Abbreviated
Profile
of
Hearing
Aid
Benefit
Speech
Reception
Thresholds
at
50%
Mean
Acceptable
Noise
Levels
Scale
Mean
SD
Ease
of
communication
25.0
15.6
0.48
0.04*
0.31
0.21
Background
noise
51.7
21.2
0.47
0.05*
0.46
0.05*
Reverberation
60.8
20.3
0.43
0.07
0.51
0.03*
Aversiveness
31.6
21.1
0.40
0.10
0.36
0.14
Global
scale
45.8
17.5
0.46
0.03*
0.47
0.05*
The
columns
on
the
right
side
give
Pearson
correlation
coefficients
examining
the
relationships
between
APHAB
scales,
Speech
Reception
Thresholds
at
50%,
and
mean
ANL
values.
significant
on
the
0.05
level.
ANL,
acceptable
noise
level;
APHAB,
abbreviated
profile
of
hearing
aid
benefit.
364
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
significantly
related
to
a
higher
score
on
the
background
noise
scale.
Furthermore,
lower
SRT50%
values
were
significantly
associated
with
greater
"ease
of
communication"
scale.
Higher
ANL
values
correlated
significantly
with
more
perceived
prob-
lems
in
reverberant
environments.
The
percent
correct
score
at
SNR
40
dB
is
not
related
to
any
APHAB
scale.
DISCUSSION
Effect
of
NRA
on
ANL
This
study
has
shown
that
the
clinically
available
single-
microphone
NRA
ClearVoice
leads
to
a
significantly
higher
acceptance
of
background
noise
among
CI
users.
The
observed
3.6
dB
improvement
in
ANL
due
to
the
NRA
was
comparable
with
improvements
found
in
studies
of
noise
reduction
effects
in
hearing
aid
users
(Mueller
et
al.
2006;
Peeters
et
al.
2009;
Pisa
et
al.
2010).
However,
in
contrast
with
the
ANL
results
of
this
study,
Holden
et
al.
(2013)
did
not
observe
significant
group
mean
differences
in
ANL
due
to
ClearVoice
in
CI
users.
We
suggest
that
this
difference
could
be
explained
by
the
use
of
different
noise
types.
Holden
et
al.
used
a
12-talker
babble
noise,
whereas
we
used
a
steady
state
speech-spectrum
noise.
A
babble
noise
contains
some
modulations,
and
this
may
have
decreased
the
effect
of
the
NRA
in
Holden's
study.
The
NRA
attenuates
the
speech
and
noise
only
if
noise
is
detected.
This
is
confirmed
by
the
finding
that
the
mean
of
the
MCL
did
not
change
for
the
NRA-off
versus
the
NRA-on
conditions.
Remarkably,
the
effect
of
the
NRA
on
ANL
is
not
significantly
related
to
the
ANL
signal
to
noise
ratio.
It
is
possible
that
listeners
use
other
criteria
than
the
overall
signal
to
noise
ratio,
which
is
directly
related
to
speech
intelligibility,
making
the
relationship
between
the
signal
to
noise
ratio
and
ANL
benefit
less
clear.
For
example,
the
loudness
of
the
noise
in
gaps
between
words
and
between
sentences
could
serve
as
a
criterion.
For
these
gaps,
the
momentary
signal
to
noise
ratio
is
low
and
is
independent
of
the
overall
signal
to
noise
ratio.
The
attenuation
in
the
gaps
is
there-
fore
independent
of
the
overall
signal
to
noise
ratio
as
well.
At
this
point,
we
wonder
what
criteria
a
listener
uses
in
deter-
mining
his
or
her
ANL.
It
is
clear
from
our
study
that
speech
intelligibility
is
not
the
primary
criterion
because
ANL
scores
improved
due
to
noise
reduction,
but
speech
intelligibility
at
similar
SNR
levels
did
not.
Previous
studies
indicated
that
ANL
scores
are
not
related
to
the
SRT
in
noise
(SRT50%)
(Nabelek
et
al.
2004;
Mueller
et
al.
2006;
Plyler
et
al.
2008;
Peeters
et
al.
2009;).
In
most
of
these
studies,
the
speech
material
of
the
ANL
differed
from
the
speech
material
of
the
speech-in-noise
test.
We
used
the
same
speech
files
and
noise
files
but
still
did
not
find
a
significant
correlation
between
ANL
and
speech
intelligibility.
Nevertheless,
although
speech
intelligibility
is
not
the
primary
criterion,
our
data
suggest
that
it
also
played
a
role.
We
were
able
to
calculate
the
word
score
at
the
ANL
signal
to
noise
ratio
from
the
results
of
the
speech-in-noise
test
for
each
participant.
Results
(Fig.
2)
showed
that
the
vast
majority
of
word
scores
at
the
ANL
SNR
were
above
50%,
although
participants
used
different
intel-
ligibility
criteria
in
the
BNL
measurement.
The
instruction
given
to
the
participants
with
regard
to
establishing
BNL
was
"select
the
level
of
the
background
noise
that
you
would
be
willing
to
accept
or
'put-up-with'
without
becoming
tense
and
tired
while
following
the
story."
Perhaps
participants
differ
in
the
weight
they
give
to
the
phrase
"while
following
the
story."
Furthermore,
the
listeners'
perception
of
their
own
ability
to
follow
the
speech
could
lead
to
over-
or
underestimations
of
the
true
intelligibility
percentage
(Saunders
&
Cienkowski
2002).
We
hypothesize
that
participants
made
ANL
judgments
based
on
the
loudness
of
the
noise
in
the
gaps
between
words
and
sentences,
as
argued
in
the
previous
para-
graph,
in
combination
with
the
less
important
intelligibility
crite-
rion
that
provides
a
ceiling
effect
for
ANL
values
as
Mueller
et
al.
have
suggested
(Mueller
et
al.
2006).
They
argued
that
the
listener
might
shift
from
a
criterion
based
on
loudness
of
the
noise
in
the
gaps
to
a
speech
intelligibility
criterion
if
the
background
noise
is
raised
to
such
a
level
that
speech
perception
is
degraded.
Effect
of
NRA
on
Speech
Intelligibility
in
Noise
Although
previous
studies
have
found
improvements
in
speech
intelligibility
in
steady
state
speech-spectrum
noise
(Buechner
et
al.
2010;
Advanced
Bionics
2012a;
Kam
et
al.
2012),
our
study
could
not
demonstrate
a
significant
benefit.
This
is
in
accordance
with
the
findings
of
Holden
et
al.
(2013)
for
steady
state
noise.
Several
factors
might
have
contributed
to
the
differences
in
findings
of
the
ClearVoice
studies
men-
tioned.
First,
all
studies
used
a
small
number
of
participants,
which
increased
changes
of
unrepresentative
samples
and
of
the
occurrence
of
false-positive
and
false-negative
results.
This
study
had
the
statistical
power
to
detect
a
difference
in
speech
intelligibility
measures
between
NRA-on
and
NRA-off
condi-
tions
of
0.65
dB
for
the
SRT50%
measure
and
1.3
dB
for
the
SRT70%
measure.
Given
a
mean
slope
of
6.4%/dB
at
50%
intelligibility,
a
difference
in
word
score
of
Z4.2%
could
be
detected.
This
study
thus
had
the
statistical
power
to
detect
a
clinically
significant
difference
in
SRT
measures.
Second,
we
considered
the
signal
to
noise
ratio.
All
studies
used
SNRs
in
the
range
of
0
to
10
dB,
and
we
used
SNRs
of
0
to
5
dB
for
the
majority
of
participants.
So
it
is
not
likely
that
the
SNR
would
have
been
a
reason
for
the
different
findings
between
the
studies.
Third,
different
speech
and
noise
levels
were
reported.
Buechner
et
al.
used
a
noise
level
of
55
dB,
whereas
Holden et
al.
reported
a
speech
level
of
50
dB
(SPL)
for
the
condition
with
speech-shaped
noise.
Kam
et
al.
used
70
dB
(SPL)
noise
level,
and
our
study
used
a
70
dB
(SPL)
speech
level.
Results
for
soft
speech
of
50
to
60
dB
(SPL)
depend
on
the
IDR
setting.
An
IDR
of
60
dB
or
more
is
required
for
maximum
speech
intelligibil-
ity
in
quiet
conditions
for
these
soft
speech
levels
(Spahr
et
al.
2007).
Holden
and
colleagues
reported
IDR
settings
of
60
dB
or
more
for
all
but
one
of
the
participants.
In
a
roving
level
speech-
in-noise
test,
Haumann
et
al.
(2010)
did
not
find
any
difference
in
SRT
for
Advanced
Bionics
Harmony
CI
if
they
added
a
50
dB
level
into
their
test.
So
it
is
not
likely
that
level
is
a
reason
for
the
different
findings
between
studies,
either,
provided
the
IDR
of
60
dB.
Fourth,
the
studies
that
reported
volume
adjust-
ments
or
T-
and
M-level
changes
in
the
majority
of
participants
reported
the
best
improvements
due
to
ClearVoice
(Buechner
et
al.
2010;
Schramm
et
al.
2011;
Noël-Petroff
et
al.
2013).
These
adjustments
increase
the
level
of
the
signal
and
alter
the
slope
of
the
input-output
mapping
function
of
the
CI.
We
do
not
expect
that
these
changes
alone
have
any
effect
on
the
SRT,
provided
that
the
IDR
setting
is
large
enough
(also
see
Haumann
et
al.
2010;
Spahr
et
al.
2007).
A
combined
effect
from
an
NRA
and
an
increase
in
volume
or
M
levels
is
more
likely.
An
NRA
atten-
uates
the
noisy
parts
of
the
signals
but
leaves
out
the
speech-
dominated
peaks
of
the
signals.
An
increase
in
the
volume
or
M
level
means
an
increase
in
the
slope
of
the
input-output
mapping
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,357-367
365
function,
which
gives
a
further
enhancement
of
the
higher
level
speech-dominated
peaks.
Further
research
is
needed
to
investi-
gate
this
possible
interaction
between
the
effect
of
an
NRA
and
an
increase
in
M
levels.
A
possible
explanation
for
the
lack
of
benefit
in
speech
intelligibility
measures
in
our
study
is
that
the
NRA
may
have
introduced
distortions
of
the
speech
signal
in
the
SNR
range
used.
The
steady
state
speech-spectrum
noise
was
pre-
sented
3
sec
before
the
start
of
a
sentence.
This
enabled
the
NRA
to
make
an
optimal
estimate
of
the
noise
spectrum.
But
the
instantaneous
SNR
may
have
been
under-
or
over-
estimated,
which
may
have
resulted
in
the
application
of
a
wrong
gain.
That
would
then
have
caused
nonrelevant
sto-
chastic
fluctuations
in
the
signal
envelope,
which
can
be
det-
rimental
to
speech
perception
(Dubbelboer
&
Houtgast
2007;
Kim
&
Loizou
2011;
Loizou
&
Kim
2011).
Qazi
et
al.
(2013)
reported
that
clear,
low-frequency
modulations
in
time
and
frequency
seem
to
be
the
most
important
factor
for
preserving
speech
intelligibility.
As
long
as
the
presentation
of
speech
maxima
remains
ideal,
CI
subjects
can
tolerate
very
high
lev-
els
of
distortions
in
the
speech
segments.
Based
on
this
obser-
vation,
we
suggest
that
ClearVoice
may
give
distortions
in
the
low-frequency
modulations.
A
higher
threshold
for
the
gain
function
could
perhaps
improve
the
low-frequency
modula-
tions.
This
is
in-line
with
Mauger
et
al.
(2012),
who
demon-
strated
that
a
positive
gain
function
threshold
provides
more
noise
reduction
and
gives
the
best
improvement
of
speech
understanding
in
noise
for
CI
subjects.
Effect
of
NRA
in
Relation
to
SR
Thresholds
A
second
purpose
of
the
study
was
to
test
the
hypothesis
that
the
effect
of
a
single-microphone
NRA
correlates
to
the
spectral
resolution
of
the
CI
stimulation.
We
did
not
find
a
cor-
relation
between
SR
discrimination
thresholds
and
the
benefit
of
the
NRA
for
the
ANL.
We
argued
earlier
that
ANL
values
are
not
primarily
based
on
intelligibility,
but
more
likely
on
the
noise
in
gaps
between
words
and
sentences.
It
is
not
likely
that
a
better
spectral
resolution
leads
to
a
different
loudness
percep-
tion
of
the
noise
in
the
gaps.
This
could
explain
the
absence
of
a
relationship
between
ANL
benefit
due
to
the
NRA
and
SR
discrimination
thresholds.
We
did
not
find
a
significant
corre-
lation
between
SR
thresholds
and
the
difference
between
the
speech
measures
for
the
NRA-on
and
NRA-off
conditions.
So,
for
speech
measures
our
hypothesis
was
not
confirmed.
Next,
we
looked
at
the
relationship
between
SR
discrimina-
tion
thresholds
and
speech-in-noise
measures.
We
did
not
find
a
significant
correlation
between
the
SRT50%
and
the
SR
dis-
crimination
threshold,
although
a
trend
was seen
that
better
SR
thresholds
are
associated
with
better
SRTs.
Our
results
are
in
accordance
with
the
results
reported
by
Anderson
et
al.
(2011)
but
are
in
contrast
with
the
results
of
Won
et
al.
(2007)
who
reported
a
correlation
between
SR
scores
and
word
recognition
in
noise.
A
possible
explanation
for
this
discrepancy
is
that
Won
et
al.
used
individual
words
in
noise,
whereas
Anderson
et
al.
and
our
study
used
word
scoring
for
a
sentence-in-noise
test.
Due
to
the
use
of
contextual
information,
the
intelligibility
of
words
in
sentences
is
influenced
more
by
linguistic
and
cogni-
tive
factors
than
by
the
understanding
of
individual
words.
It
can
be
assumed
that
these
factors
added
more
variance
to
the
data
than
the
differences
in
spectral
resolution
did.
For
the
quiet
condition
(scores
at
SNR
40
dB),
we
found
that
words
in
sentences
and
SR
discrimination
thresholds
were
significantly
correlated.
This
is
in
agreement
with
the
studies
of
Anderson
et
al.
(2011)
and
Won
et
al.
(2007).
NRA
and
Self-Perceived
Communication
Problems
An
important
question
is
whether
the
outcome
measures
of
ANL
and
speech
intelligibility
in
noise can
be
extrapolated
to
communication
problems
in
daily
life
as
measured
with
the
APHAB
questionnaire.
Results
indicated
that
CI
users
with
lower
ANL
values
and
better
speech
intelligibility
in
noise
reported
significantly
fewer
problems
in
daily
life
on
the
APHAB
overall
scale
and
several
subscales.
This
confirms
that
the
outcome
measures
we
have
chosen
were
related
to
everyday
life
communication
and
therefore
justify
the
use
of
the
ANL
and
SRT
in
evaluating
a
NRA.
The
fact
that
correlations
were
only
modest
shows
that
also
other
factors
have
an
effect
on
daily
communication.
It
is
likely
that
CI
users
use
nonauditory
infor-
mation,
for
example
visual
cues
obtained
from
lipreading.
The
APHAB
scores
showed
that
the
number
of
perceived
problems
was
greatest
for
the
"reverberation"
scale
and
fewest
for
the
"ease
of
communication"
scale.
This
is
consistent
with
previ-
ously
reported
APHAB
scores
of
CI
users
(Plyler
et
al.
2008;
Donaldson
et
al.
2009).
In
agreement
with
Donaldson
et
al.,
our
results
showed
a
significant
correlation
between
ANL
val-
ues
and
APHAB
scores,
although
our
correlations
coefficients
were
somewhat
lower.
The
"aversiveness
to
sound"
scale
of
the
APHAB
was
not
significant
in
relation
to
ANL
values.
This
is
not
surprising
because
the
questions
of
this
scale
concerned
loud,
nonspeech
sounds,
instead
of
noise
during
speech
percep-
tion.
We
fitted
a
linear
equation
to
the
data
of
SRT50%
and
the
APHAB
Global
score.
The
slope
indicates
that
an
improvement
of
1
dB
in
the
SNR
gives
an
improvement
of
2.5%
in
APHAP
Global
score.
A
fit
of
ANL
data
with
the
APHAB
Global
score
indicates
that
1
dB
of
ANL
improvement
gives
an
APHAB
improvement
of
1.7%.
The
mean
improvement
of
3.6
dB
for
ANL
due
to
the
NRA
means
a
6.1%
reduction
in
reported
prob-
lems
with
the
APHAB
questionnaire,
which
is
a
modest
but
rel-
evant
reduction
in
perceived
communication
problems.
General
Discussion
and
Conclusions
We
conclude
that
the
NRA
ClearVoice
improves
listening
comfort
for
CI
users,
in
the
sense
that
they
can
tolerate
a
higher
noise
level
when
listening
to
speech
in
background
noise.
The
results
of
the
APHAP
questionnaire
suggest
that
the
improved
noise
tolerance
leads
to
fewer
complaints
in
everyday
listening
situations.
The
improvement
of
listening
comfort
in
steady
state
noise
due
to
a
single-microphone
NRA
for
CI
users
is
in
accor-
dance
with
findings
for
NRAs
in
hearing
aids.
Speech
intelligibility
in
noise
remained unchanged
by
the
NRA
in
this
study.
This
study
at
least
supports
the
idea
that
in
clinical
CI
applications
NRAs
contribute
more
to
the
improve-
ment
of
listening
comfort
than
to
the
improvement
of
speech
understanding
in
noise.
Our
hypothesis
that
CI
recipients
with
lower
spectral
resolu-
tion
might
have
more
benefit
from
noise
reduction
than
CI
users
with
higher
spectral
resolution
holds
neither
for
noise tolerance
nor
for
speech
intelligibility
in
noise.
The
improvement
of
noise
tolerance
is
not
related
to
SR
discrimination
thresholds,
speech
intelligibility
measures,
or
signal
to
noise
ratio
in
this
study.
366
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
Furthermore,
SR
discrimination
thresholds
are
not
related
to
the
effect
of
ClearVoice
on
speech
intelligibility
in
noise
or
to
the
speech
intelligibility
in
noise
ratios.
Maybe,
other
nonauditory
factors,
such
as
linguistic
and
cognitive
factors,
add
more
vari-
ance
to
the
speech
understanding
in
noise
and
noise
tolerance
than
spectral
resolution
does.
This
is
a
topic
for
further
research.
ACKNOWLEDGMENTS
The
authors
gratefully
acknowledge
the
participation
of
the
research
sub-
jects,
and
they
thank
Michael
Brocaar,
Agnes
Doorduin,
Maarten
Meijer,
and
Fanny
Scherf
for
facilitating
some
of
the
experimental
work
and
data
collection.
This
work
was
supported
by
Advanced
Bionics.
Portions
of
the
data
were
presented
at
the
11th
European
Symposium
on
Pediatric
Cochlear
Implantation,
May
23-26,2013,
Istanbul.
The
authors
declare
no
other
conflict
of
interest.
Address
for
correspondence:
J.
Gertjan
Dingemanse,
Erasmus
Medical
Center,
Department
of
ENT,
Room
D126,
P.O.
Box
2040,
3000
CA
Rotterdam,
The
Netherlands.
E-mail:
g.dingemanse@erasmusmc.n1
Received
January
30,
2014;
accepted
October
7,2014.
REFERENCES
Advanced
Bionics.
(2012a).
ClearVoice.
Clinical
Results.
Valencia
CA:
Advanced
Bionics.
Advanced
Bionics.
(2012b).
ClearVoice.
Technical
Facts.
Valencia
CA:
Advanced
Bionics.
Anderson,
E.
S.,
Nelson,
D.
A.,
Kreft,
H.,
et
al.
(2011).
Comparing
spa-
tial
tuning
curves,
spectral
ripple
resolution,
and
speech
perception
in
cochlear
implant
users.
JAcoust
Soc
Am,
130,
364-375.
Azadpour,
M.,
&
McKay,
C.
M.
(2012).
A
psychophysical
method
for
mea-
suring
spatial
resolution
in
cochlear
implants.
J
Assoc
Res
Otolwyngol,
13,
145-157.
Bentler,
R.,
Wu,
Y.
H.,
Kettel,
J.,
et
al.
(2008).
Digital
noise
reduction:
Out-
comes
from
laboratory
and
field
studies.
Int
JAudiol,
47,447-460.
Buechner,
A.,
Brendel,
M.,
Saalfeld,
H.,
et
al.
(2010).
Results
of
a
pilot
study
with
a
signal
enhancement
algorithm
for
HiRes
120
cochlear
implant
users.
Otol
Neurotol,
31,
1386-1390.
Chung,
K.,
Zeng,
F.
G.,
Acker,
K.
N.
(2006).
Effects
of
directional
micro-
phone
and
adaptive
multichannel
noise
reduction
algorithm
on
cochlear
implant
performance.
JAcoust
Soc
Am,
120,
2216-2227.
Chung,
K.,
Nelson,
L.,
Teske,
M.
(2012).
Noise
reduction
technologies
implemented
in
head-worn
preprocessors
for
improving
cochlear
implant
performance
in
reverberant
noise
fields.
Hear
Res,
291,
41-51.
Cox,
R.
M.,
&
Alexander,
G.
C.
(1995).
The
abbreviated
profile
of
hearing
aid
benefit.
Ear
Hear,
16,
176-186.
Dawson,
P.
W,
Mauger,
S.
J.,
Hersbach,
A. A.
(2011).
Clinical
evaluation
of
signal-to-noise
ratio-based
noise
reduction
in
Nucleus®
cochlear
implant
recipients.
Ear
Hear,
32,
382-390.
Donaldson,
G.
S.,
Chisolm,
T.
H.,
Blasco,
G.
P.,
et
al.
(2009).
BKB-
SIN
and
ANL
predict
perceived
communication
ability
in
cochlear
implant
users.
Ear
Hear,
30,
401-410.
Dubbelboer,
F.,
&
Houtgast,
T.
(2007).
A
detailed
study
on
the
effects
of
noise
on
speech
intelligibility.
JAcoust
Soc
Am,
122,
2865-2871.
Haumann,
S.,
Lenarz,
T.,
Buchner,
A.
(2010).
Speech
perception
with
cochlear
implants
as
measured
using
a
roving-level
adaptive
test
method.
ORL
J
Otorhinolaryngol
Relat
Spec,
72,
312-318.
Henry,
B.
A.,
&
Turner,
C.
W.
(2003).
The
resolution
of
complex
spectral
patterns
by
cochlear
implant
and
normal-hearing
listeners.
JAcoust
Soc
Am,
113,
2861-2873.
Henry,
B.
A.,
Turner,
C.
W,
Behrens,
A.
(2005).
Spectral
peak
resolution
and
speech
recognition
in
quiet:
Normal
hearing,
hearing
impaired,
and
cochlear
implant
listeners.
JAcoust
Soc
Am,
118,
1111-1121.
Hersbach,
A.
A.,
Arora,
K.,
Mauger,
S.
J.,
et
al.
(2012).
Combining
direc-
tional
microphone
and
single-channel
noise
reduction
algorithms:
A
clinical
evaluation
in
difficult
listening
conditions
with
cochlear
implant
users.
Ear
Hear,
33,
e13—e23.
Hochberg,
I.,
Boothroyd,
A.,
Weiss,
M.,
et
al.
(1992).
Effects
of
noise
and
noise
suppression
on
speech
perception
by
cochlear
implant
users.
Ear
Hear,
13,
263-271.
Holden,
L.
K.,
Brenner,
C.,
Reeder,
R.
M.,
et
al.
(2013).
Postlingual
adult
performance
in
noise
with
HiRes
120
and
ClearVoice
Low,
Medium,
and
High.
Cochlear
Implants
Int,
14,
276-286.
Hu,
Y.,
Loizou,
P.
C.,
Li,
N.,
et
al.
(2007).
Use
of
a
sigmoidal-shaped
func-
tion
for
noise
attenuation
in
cochlear
implants.
JAcoust
Soc
Am,
122,
EL128—EL134.
Jones,
G.
L.,
Won,
J.
H.,
Drennan,
W.
R.,
et
al.
(2013).
Relationship
between
channel
interaction
and
spectral-ripple
discrimination
in
cochlear
implant
users.
JAcoust
Soc
Am,
133,
425-433.
Kam,
A.
C.,
Ng,
I.
H.,
Cheng,
M.
M.,
et
al.
(2012).
Evaluation
of
the
Clear-
Voice
Strategy
in
Adults
Using
HiResolution
Fidelity
120
Sound
Pro-
cessing.
Clin
Exp
Otorhinolaryngol,
5(Suppl
1),
S89—S92.
Kasturi,
K.,
&
Loizou,
P.
C.
(2007).
Use
of
S-shaped
input-output
functions
for
noise
suppression
in
cochlear
implants.
Ear
Hear,
28,
402-411.
Kim,
G.,
&
Loizou,
P.
C.
(2011).
Gain-induced
speech
distortions
and
the
absence
of
intelligibility
benefit
with
existing
noise-reduction
algo-
rithms.
JAcoust
Soc
Am,
130,
1581-1596.
Kokkinakis,
K.,
Azimi,
B.,
Hu,
Y.,
et
al.
(2012).
Single
and
multiple
micro-
phone
noise
reduction
strategies
in
cochlear
implants.
Trends
Amplif,
16,
102-116.
Kushner,
H.
J.,
&
Yin,
G.
(2003).
Stochastic
Approximation
and
Recursive
Algorithms
and
Applications.
New
York:
Springer-Verlag.
Litvak,
L.
M.,
Spahr,
A.
J.,
Saoji,
A.
A.,
et
al.
(2007).
Relationship
between
perception
of
spectral
ripple
and
speech
recognition
in
cochlear
implant
and
vocoder
listeners.
JAcoust
Soc
Am,
122,
982-991.
Loizou,
P.
C.,
&
Kim,
G.
(2011).
Reasons
why
current
speech-enhancement
algorithms
do
not
improve
speech
intelligibility
and
suggested
solutions.
IEEE
Trans
Audio
Speech
Lang
Processing,
19,
47-56.
Luts,
H.,
Eneman,
K.,
Wouters,
J.,
et
al.
(2010).
Multicenter
evaluation
of
signal
enhancement
algorithms
for
hearing
aids.
JAcoust
Soc
Am,
127,
1491-1505.
Mauger,
S.
J.,
Dawson,
P.
W,
Hersbach,
A. A.
(2012).
Perceptually
opti-
mized
gain
function
for
cochlear
implant
signal-to-noise
ratio
based
noise
reduction.
JAcoust
Soc
Am,
131,
327-336.
Mueller,
H.
G.,
Weber,
J.,
Hornsby,
B.
W.
(2006).
The
effects
of
digital
noise
reduction
on
the
acceptance
of
background
noise.
Trends
Amplif,
10,
83-93.
Nabelek,
A.
K.,
Freyaldenhoven,
M.
C.,
Tampas,
J.
W,
et
al.
(2006).
Accept-
able
noise
level
as
a
predictor
of
hearing
aid
use.
JAm
Acad
Audiol,
17,
626-639.
Nabelek,
A.
K.,
Tampas,
J.
W,
Burchfield,
S.
B.
(2004).
Comparison
of
speech
perception
in
background
noise
with
acceptance
of
background
noise
in
aided
and
unaided
conditions.
J
Speech
Lang
Hear
Res,
47,
1001-1011.
Nabelek,
A.
K.,
Tucker,
F.
M.,
Letowski,
T.
R.
(1991).
Toleration
of
back-
ground
noises:
Relationship
with
patterns
of
hearing
aid
use
by
elderly
persons.
J
Speech
Hear
Res,
34,
679-685.
Noel-Petroff,
N.,
Mathias,
N.,
Uhnann,
C.,
et
al.
(2013).
Pediatric
evalua-
tion
of
the
ClearVoiceTM
speech
enhancement
algorithm
in
everyday
life.
Audiology
Research,
3,
e9.
Peeters,
H.,
Kuk,
F.,
Lau,
C.
C.,
et
al.
(2009).
Subjective
and
objective
evaluation
of
noise
management
algorithms.
JAm
Acad
Audiol,
20,
89-98.
Pisa,
J.,
Burk,
M.,
Galster,
E.
(2010).
Evidence-based
design
of
a
noise-
management
algorithm.
Hearing
J,
63,
42-44.
Plyler,
P.
N.,
Bating,
J.,
von
Hapsburg,
D.
(2008).
The
acceptance
of
back-
ground
noise
in
adult
cochlear
implant
users.
J
Speech
Lang
Hear
Res,
51,
502-515.
Qazi,
0.
U.,
van
Dijk,
B.,
Moonen,
M.,
et
al.
(2013).
Understanding
the
effect
of
noise
on
electrical
stimulation
sequences
in
cochlear
implants
and
its
impact
on
speech
intelligibility.
Hear
Res,
299,
79-87.
Ricketts,
T.
A.,
&
Hornsby,
B.
W.
(2005).
Sound
quality
measures
for
speech
in
noise
through
a
commercial
hearing
aid
implementing
digital
noise
reduction.
JAm
Acad
Audiol,
16,
270-277.
Robbins,
H.,
&
Monro,
S.
(1951).
A
stochastic
approximation
method.
Ann
Math
Stat,
22,
400-407.
Saunders,
G.
H.,
&
Cienkowski,
K.
M.
(2002).
A
test
to
measure
subjective
and
objective
speech
intelligibility.
J
Am
Acad
Audiol,
13,
38-49.
Schramm,
D.,
Pickard,
E.
S.,
Beauregard,
Y.,
et
al.
(2011).
F034
Evalua-
tion
of
the
ClearVoiceTM
strategy
in
children
using
HiResolution
Fidelity
120®
sound
processing.
Int
J
Pediatr
Otorhinol,
75,89.
Spahr,
A.
J.,
Dorman,
M.
F.,
Loiselle,
L.
H.
(2007).
Performance
of
patients
using
different
cochlear
implant
systems:
Effects
of
input
dynamic
range.
Ear
Hear,
28,
260-275.
DINGEMANSE
&
GOEDEGEBURE
/
EAR
&
HEARING,
VOL.
36,
NO.
3,
357-367
367
Spriet,
A.,
Van
Deun,
L.,
Eftaxiadis,
K.,
et
al.
(2007).
Speech
understanding
in
background
noise
with
the
two-microphone
adaptive
beamformer
BEAM
in
the
Nucleus
Freedom
Cochlear
Implant
System.
Ear
Hear,
28,
62-72.
Studebaker,
G.
A.
(1985).
A
rationalized
arcsine
transform.
J
Speech
Lang
Hear
Res,
28,
455-462.
Toledo,
F,
Loizou,
P.,
Lobo,
A.
(2003).
Subspace
and
envelope
subtraction
algo-
rithms
for
noise
reduction
in
cochlear
implants.
In
Engineering
in
Medicine
and
Biology
Society,
2003.
Proceedings
of
the
25th
Annual
International
Conference
of
the IEEE
(Vol.
3,
pp.
2002-2005).
Piscataway,
NJ:
IEEE
.
Versfeld,
N.
J.,
Daalder,
L.,
Festen,
J.
M.,
et
al.
(2000).
Method
for
the
selec-
tion
of
sentence
materials
for
efficient
measurement
of
the
speech
recep-
tion
threshold.
JAcoust
Soc
Am,
107,
1671-1684.
Won,
J.
H.,
Drennan,
W.
R.,
Rubinstein,
J.
T.
(2007).
Spectral-ripple
resolu-
tion
correlates
with
speech
reception
in
noise
in
cochlear
implant
users.
J
Assoc
Res
Otolaryngol,
8,
384-392.
Won,
J.
H.,
Jones,
G.
L.,
Drennan,
W.
R.,
et
al.
(2011).
Evidence
of
across-
channel
processing
for
spectral-ripple
discrimination
in
cochlear
implant
listeners.
JAcoust
Soc
Am,
130,
2088-2097.
Yang,
L.
P.,
&
Fu,
Q.
J.
(2005).
Spectral
subtraction-based
speech
enhance-
ment
for
cochlear
implant
patients
in
background
noise.
JAcoust
Soc
Am,
117(3
Pt
1),
1001-1004.
Zakis,
J.
A.,
Hau,
J.,
Blarney,
P.
J.
(2009).
Environmental
noise
reduction
configuration:
Effects
on
preferences,
satisfaction,
and
speech
under-
standing.
Int
J
Audiol,
48,
853-867.