Effects of a transient noise reduction algorithm on speech intelligibility in noise, noise tolerance and perceived annoyance in cochlear implant users

Dingemanse, J.Gertjan.; Vroegop, J.L.; Goedegebure, Aé.

International Journal of Audiology 57(5): 360-369

2018

Label="OBJECTIVE">To evaluate the validity and efficacy of a transient noise reduction algorithm (TNR) in cochlear implant processing and the interaction of TNR with a continuous noise reduction algorithm (CNR).Label="DESIGN">We studied the effects of TNR and CNR on the perception of realistic sound samples with transients, using subjective ratings of annoyance, a speech-in-noise test and a noise tolerance test.Label="STUDY SAMPLE">Participants were 16 experienced cochlear implant recipients wearing an Advanced Bionics Naida Q70 processor.Label="RESULTS">CI users rated sounds with transients as moderately annoying. Annoyance was slightly, but significantly reduced by TNR. Transients caused a large decrease in speech intelligibility in noise and a moderate decrease in noise tolerance, measured on the Acceptable Noise Level test. The TNR had no significant effect on noise tolerance or on speech intelligibility in noise. The combined application of TNR and CNR did not result in interactions.Label="CONCLUSIONS">The TNR algorithm was effective in reducing annoyance from transient sounds, but was not able to prevent a decreasing effect of transients on speech understanding in noise and noise tolerance. TNR did not reduce the beneficial effect of CNR on speech intelligibility in noise, but no cumulated improvement was found either.

O
Taylor
&
Francis
Taylor
&
Francis
Group
International
Journal
of
Audiology
2018;
Early
Online:
1-10
International
Journal
of
Audiology
Original
Article
Effects
of
a
transient
noise
reduction
algorithm
on
speech
intelligibility
in
noise,
noise
tolerance
and
perceived
annoyance
in
cochlear
implant
users
J.
Gertjan
Dingemanse
Jantien
L.
Vroegop,
and
Andre
Goedegebure
Department
of
ENT,
Erasmus
Medical
Center,
Rotterdam,
The
Netherlands
Abstract
Objective:
To
evaluate
the
validity
and
efficacy
of
a
transient
noise
reduction
algorithm
(TNR)
in
cochlear
implant
processing
and
the
interaction
of
TNR
with
a
continuous
noise
reduction
algorithm
(CNR).
Design:
We
studied
the
effects
of
TNR
and
CNR
on
the
perception
of
realistic
sound
samples
with
transients,
using
subjective
ratings
of
annoyance,
a
speech-in-noise
test
and
a
noise
tolerance
test.
Study
sample:
Participants
were
16
experienced
cochlear
implant
recipients
wearing
an
Advanced
Bionics
Naida
Q70
processor.
Results:
CI
users
rated
sounds
with
transients
as
moderately
annoying.
Annoyance
was
slightly,
but
significantly
reduced
by
TNR.
Transients
caused
a
large
decrease
in
speech
intelligibility
in
noise
and
a
moderate
decrease
in
noise
tolerance,
measured
on
the
Acceptable
Noise
Level
test.
The
TNR
had
no
significant
effect
on
noise
tolerance
or
on
speech
intelligibility
in
noise.
The
combined
application
of
TNR
and
CNR
did
not
result
in
interactions.
Conclusions:
The
TNR
algorithm
was
effective
in
reducing
annoyance
from
transient
sounds,
but
was
not
able
to
prevent
a
decreasing
effect
of
transients
on
speech
understanding
in
noise
and
noise
tolerance.
TNR
did
not
reduce
the
beneficial
effect
of
CNR
on
speech
intelligibility
in
noise,
but
no
cumulated
improvement
was
found
either.
Key
Words:
Cochlear
implant,
maximum
comfort
level,
ClearVoice,
SoundRelax,
transients,
transient
noise
reduction
algorithm,
acceptable
noise
level,
speech
reception
threshold,
sound
annoyance
The
British
Society
of
Audiology
The
International
Society
of
Audiology
CHAS
NORDIC
ALIDIOLOGCAL
SOCE1Y
Introduction
The
focus
of
a
Cochlear
Implant
(CI)
fitting
is
usually
on
achieving
good
speech
intelligibility.
However,
it
is
also
important
to
consider
aspects
of
listening
comfort
and
sound
quality,
especially
in
noisy
environments
(Mertens
et
al.
2015).
In
everyday
life,
people
experience
a
variety
of
sounds
that
differ
in
their
spectro-temporal
characteristics,
duration
or
loudness
and
can
be
perceived
as
disturbing,
especially
when
listening
to
speech.
Nowadays,
direc-
tional
microphones
and
single-microphone
noise
reduction
algo-
rithms
are
applied
in
CI
processors
to
reduce
the
effect
of
background
noises.
The
single-microphone
noise
reduction
is
sometimes
named
as
continuous
noise
reduction
(CNR),
because
it
is
mainly
effective
for
noises
with
a
continuous
temporal
behaviour.
Transient
sounds,
however,
will
not
be
affected
by
CNR.
Transient
sounds
are
characterised
by
a
very
fast
onset
to
the
peak
in
sound
pressure
level
(within
a
few
milliseconds),
a
fast
decay
and
a
short
duration
(from
tens
of
milliseconds
up
to
one
second).
The
peak
sound
pressure
level
of
the
transient
is
well
above
the
average
sound
pressure
level.
Korhonen
et
al.
(2013)
reported
sound
pressure
levels
and
rise
times
for
different
recorded
transients.
The
levels
varied
from
67
dB
(A,
impulse)
for
a
clicking
pen
up
to
102
dB
(A,
impulse)
for
stacking
two water
glasses.
Rise
times
ranged
from
less
than
1
ms
up
to
4
ms.
It
is
well
known
that
hearing-aids
users
frequently
perceive
transient
sounds
as
disturbing.
Hernandez,
Chalupper,
and
Powers
(2006)
reported
that
about
one-third
of
the
annoying
background
noises
commonly
encountered
by
new
hearing
instrument
wearers
were
of
a
transient
type.
In
that
study
transients
were
defined
as
noises
with
a
duration
of
<1
s.
A
fast
onset
was
not
required.
The
perceived
annoyance
of
these
transient
noises
was
slightly
lower
than
the
annoyance
of
continuous
noises,
but
still
substantial
(6.3
on
a
0-10
annoyance
rating
scale).
The
automatic gain
controls
(AGC)
of
hearing
aids
usually
use
a
fast-acting
system
to
cope
with
transient
sounds,
but
for
transients
with
a
very
fast
onset
the
AGC
is
often
too
slow.
Hence
transient
noise
reduction
(TNR)
systems
have
Correspondence:
J.
Gertjan
Dingemanse,
Department
of
ENT,
Erasmus
Medical
Center,
room
D126,
PObox
2040,
Rotterdam,
3000
CA,
The
Netherlands.
E-mail:
[email protected]
(Received
13
July
2017;
revised
22
December
2017;
accepted
2
January
2018)
ISSN
1499-2027
print/ISSN
1708-8186
online
©
2018
The
Author(s).
Published
by
Informa
UK
Limited,
trading
as
Taylor
&
Francis
Group.
This
is
an
Open
Access
article
distributed
under
the
terms
of
the
Creative
Commons
Attribution-NonCommercial-NoDerivatives
License
(http://creativecommons.org/licenses/by-nc-nd/4.0/),
which
permits
non-commercial
re-use,
distribution,
and
reproduction
in
any
medium,
provided
the
original
work
is
properly
cited,
and
is
not
altered,
transformed,
or
built
upon
in
any
way.
DOI:
10.1080/14992027.2018.1425004
Abbreviations
ANL
ANOVA
BNL
CI
MCL
M-level
CNR
RMS
SNR
SRTn
T-level
TNR
acceptable
noise
level
analysis
of
variance
background
noise
level
Cochlear
implant
most
comfortable
level
maximum
comfort
level
or
upper
stimulation
level
linked
to
MCL
continuous
noise
reduction
root
mean
square
signal-to-noise
ratio
or
speech-to-noise
ratio
speech
reception
threshold
in
noise
at
50%
intelligibility
stimulation
level
at
hearing
threshold
transient
noise
reduction.
2
J.
G.
Dingemanse
et
al.
been
developed
to
reduce
the
disturbing
effects
of
transient
sounds
in
hearing
aids.
Several
studies
have
evaluated
the
efficacy
of
a
TNR
in
hearing
aid
users
with
various
transient
noises
and
outcome
measures,
such
as
subjective
ratings
or
paired
comparisons
for
speech
clarity,
annoyance,
comfort,
loudness
and
speech
perception
tests
(DiGiovanni,
Davlin,
and
Nagaraj
2011;
Keidser
et
al.
2007;
Korhonen
et
al.
2013;
Liu
et
al.
2012).
The
results
of
these
studies
suggest
that
TNRs
are
most
effective
for
loud
transients
and
are
not
detrimental
for
speech
perception.
Compared
to
hearing
aids
users,
the
perceived
disturbing
effects
of
transient
sounds
are
not
necessarily
the
same
for
CI
users,
due
to
the
different
way
of
sound
processing
and
the
use
of
electric
stimulation.
However,
data
on
sound
annoyance
in
CI
users
are
scarce
and
we
were
only
aware
of
a
study
of
Mauger,
Arora,
and
Dawson
(2012).
They
described
noise
annoyance
ratings
of
CI
recipients
for
steady-state
noise,
4-talker
and
20-talker
noise
presented
together
with
speech
at
65
dB(SPL).
The
steady-state
noise
condition
was
rated
as
highly
annoying
(75/100
on
a
numberless
scale),
but
annoyance
was
substantially
reduced
by
their
noise
reduction
algorithm
(19/100).
The
babble
noise
condi-
tions
were
rated
as
moderately
annoying
(54/100
for
4-talker
noise
and
61/100
for
20-talker
noise)
and
the
ratings
were
less
influenced
by
noise reduction
(41/100
for
4-talker
noise
and
30/100
for
20-
talker
noise).
Similar
to
hearing
aids,
cochlear
implant
processors
use
an
AGC
to
keep
the
signal
within
the
electrical
dynamic
range
of
the
patient
and
to
prevent
discomfort
due
to
sudden
loud
sounds
(Vaerenberg
et
al.
2014).
In
most
CI
processors,
the
AGC
is
a
dual
time
constant
AGC,
with
both
a
fast
detector
and
a
slow
detector
(Boyle
et
al.
2009;
Khing,
Swanson,
and
Ambikairajah
2013;
Stone
et
al.
1999;
Moore,
Glasberg,
and
Stone
1991).
Stobich,
Zierhofer,
and
Hochmair
(1999)
investigated
the
effect
of
an
intense
transient
(a
"chink"
with
peak
sound
pressure
level
of
100
dB)
in
CI
users
that
used
a
CI
processor
with
a
dual
time
constant
AGC.
The
transient
was
spliced
onto
the
beginning
of
a
sentence
presented
at
85
dB
SPL.
They
found
that
the
dual
time
constant
compression
system
handled
the
transient
within
the
speech
effectively,
making
the
transients
less
detrimental
for
speech
perception.
However,
there
is
room
for
improvement,
as
the
attack
time
of
most
fast-acting
AGCs
is
3-5
ms.
This
is
still
too
slow
to
catch
the
onset
of
many
transients
and
the
amount
of
reduction
is
unlikely
to
be
sufficient
to
prevent
discomfort.
Therefore
a
TNR
have
recently
been
introduced
in
cochlear
implant
systems
that
is
capable
to
reduce
transients
with
onset-to-peak
levels
within
1
ms.
Dyballa
et
al.
(2015)
investigated
the
effect
of
a
TNR
in
CI
users
on
speech
intelligibility
in
quiet
and
in
two
types
of
transient
noise:
repetitive
hammer
blows
and
dishes
(clinking
cups
and
spoons).
The
noises
had
a
peak
level
of
90
dB
(SPL)
and
a
RMS
level
of
approximately
70
dB
(SPL).
Speech
perception
in
quiet
was
not
affected
by
the
algorithm.
The
speech
reception
threshold
in
noise
was
significantly
improved
by
0.4
dB
for
the
dishes
noise
and
1.7
dB
for
the
hammering
noise.
In
everyday
situations,
transients
may
be
mixed
with
continuous
background
noises,
for
example
in
a
kitchen
where
transients
from
clinking
bowls
or
plates
are
concurrent
with
continuous
noise
from
an
exhaust
hood.
In
such
situations,
TNR
and
CNR
may
be
activated
simultaneously
in
a
CI
processor
or
hearing
aid.
It
is
unknown
if
a
combination
of
TNR
and
CNR
has
additional
positive
or
negative
effects
on
sound
perception.
Transients
may
cause
less
functioning
of
a
CNR.
If
a
transient
sound
occurs,
the
instantaneous
SNR
estimate
of
a
CNR
algorithm
becomes
positive
(the
signal
level
is
above
the
estimated
noise
level
that
is
based
on
a
longer
time
window)
and
less
attenuation
is
applied
by
the
algorithm.
If
there
are
many
transients
the
estimated
noise
level
may
become
inaccurate.
A
TNR
may
reduce
the
high
peak
levels
and
prevent
from
less
functioning
of
the
CNR,
resulting
in
a
positive
interaction
between
CNR
and
TNR
in
conditions
where
transients
and
continuous
noises
are
mixed.
Next,
a
combination
of
TNR
and
CNR
may
reduce
the
sound
annoyance
and
increase
the
noise
tolerance
more
than
each
algorithm
alone.
As
only
limited
information
was
available
about
how
transient
sounds
are
perceived
by
CI-users
and
about
the
potential
benefit
of
TNR,
we
wanted
to
investigate
the
efficacy
of
TNR
in
CI-
users
on
speech
perception,
noise
tolerance
and
annoyance.
Our
tests
were
performed
in
a
group
of
experienced
CI
users,
using
a
subset
of
realistic
sound
recordings
with
transients
that
were
able
to
activate
the
TNR
algorithm.
Furthermore,
we
investigated
the
effect
of
these
transients
without
algorithm
to
learn
more
about
the
need
for
TNR.
We
wanted
to
answer
the
following
research
questions:
(1)
How
annoying
and
how
detrimental
for
speech
intelligibility
in
noise
are
transients
that
are
able
to
activate
a
TNR
algorithm
applied
in
CI
users?
(2)
Does
the
application
of
TNR
in
CI
users
increase
the
speech
intelligibility
in
noise,
the
noise
tolerance
and
reduce
perceived
annoyance
for
transients
in
speech
and
noise?
(3)
Does
the
combined
application
of
TNR
and
CNR
in
CI
users
result
in
a
cumulated
improvement
in
speech
intelligibility,
noise
tolerance
and
perceived
annoyance
in
noisy
backgrounds
that
contain
transient
sounds?
Materials
and
methods
Participants
Sixteen
CI
users
were
included
in
the
study,
as
indicated
by
an
a
priori
power
analysis
(see
Data
analysis).
The
sixteen
participants
ranged
in
age
from
40
to
81
years
(group
mean
66
years;
SD
=12.0).
All
participants
were
unilaterally
implanted
with
an
Advanced
Bionics
cochlear
implant
(HiRes
90K
implant).
The
average
duration
of
implant
use
was
7.4
(SD
3.7)
years
with
a
minimum
of
one
year
of
use.
All
participants
used
at
least
14
active
electrodes
and
the
HiRes
Optima-S
sound
coding
strategy.
In
the
daily
used
programme,
all
but
two
used
the
CNR
algorithm
Transient
Noise
Reduction
3
ClearVoice
and
all
but
three
did
not
use
the
TNR
algorithm
SoundRelax.
The
input
dynamic
range
(IDR)
setting
was
between
55
and
63
dB
(13
participants
had
an
IDR
of
60
dB).
Free
field
thresholds
were
better
than
40
dB
HL
(average
of
500,
1000,
2000
and
4000Hz)
for
all
participants
and
for
nine
participants
better
than
30
dB
HL.
Four
participants
wore
a
hearing
aid
in
the
non-
implanted
ear,
but
the
hearing
aid
was
switched
off
during
the
tests.
Without
hearing
aids
all
participants
had
severe
hearing
loss
of
at
least
100
dB
HI,
pure
tone
average
(PTA),
except
two
who
had
a
PTA
of
80
and
92
dB
HL.
All
participants
were
Dutch
native
speakers.
For
inclusion
in
this
study,
a
phoneme
score
of
at
least
70%
on
clinically
used
Dutch
consonant-vowel-consonant
word
lists
(Bosman
and
Smoorenburg
1995)
was
required.
Participants
were
required
to
sign
a
written
informed
consent
form
before
participating
in
the
study.
The
Erasmus
Medical
Center
Ethics
Committee
approved
the
study
protocol
for
use
with
CI
recipients.
Cochlear
implant
algorithms
The
study
used
an
Advanced
Bionics
Naida
Q70
sound
processor,
which
contains
a
TNR
algorithm
called
SoundRelax
and
a
CNR
algorithm
called
ClearVoice.
Both
are
proprietary
algorithms
of
Advanced
Bionics
(Stafa,
Switzerland).
The
TNR
algorithm
detects
transients
by
comparing
a
fast
following
envelope
and
a
slow
following
envelope
of
the
broadband
incoming
signal.
First,
the
absolute
peak
level
of
the
noise transient
(fast
envelope)
has
to
exceed
78
dB
SPL.
Second,
the
transient
has
to
rise
rapidly
above
the
slow
envelope
level
by
at
least
20
dB,
with
a
level
change
of
at
least
20dB/ms.
If
these
criteria
are
met,
the
level
of
the
transient
is
attenuated.
If
the
transient
level
is
between
20
and
26
dB
above
the
slow
envelope
level,
the
attenuation
is
14
dB
and
if
the
transient
level
is
greater
than
26
dB
above
the
slow
envelope
level,
the
attenuation
is
20
dB.
After
activation
of
the
TNR
algorithm,
the
amount
of
level
reduction
decreases
exponentially
to
zero
within
200
ms.
The
TNR
algorithm
is
designed
to
have
minimal
impact
on
the
speech
signal,
which
was
confirmed
by
a
study
of
Dyballa
et
al.
(2015).
The
TNR
acts
early
in
the
signal
processing
path,
before
the
automatic
gain
control
(AGC).
The
AGC
of
the
sound
processor
has
a
dual-time-constant
compression:
a
slow-acting
compressor
(attack
time
240
ms,
release
time
1500
ms)
becomes
active
when
the
input
level
exceeds
the
compression
threshold
of
63
dB
SPL
and
the
fast-
acting
compressor
(attack
time
3
ms,
release
time
80
ms)
becomes
active
at
a
threshold
of
71
dB
SPL,
thus
avoiding
uncomfortable
loudness.
Both
compressors
have
a
compression
ratio
of
12:1
(Boyle
et
al.
2009)
and
act
on
the
broadband
signal.
CNR
algorithm ClearVoice
has
the
aim
to
improve
overall
signal-to-noise
ratio
(SNR)
by
suppression
of
frequency
channels
lacking
useful
information
for
understanding
speech.
The
CNR
algorithm
is
applied
behind
the
AGC
and
is
active
in
the
different
frequency
channels.
Within
each
channel,
the
algorithm
calculates
a
long-term
estimation
of
the
noise
level
using
a
1.3
s
time
window
and
an
instantaneous
SNR.
Depending
on
the
difference
between
the
instantaneous
SNR
and
the
long-term
average
SNR,
a
negative
gain
is
applied.
In
this
study
we
used
the
Medium
setting
of
ClearVoice,
resulting
in
a
negative
gain
down
to
—12
dB
(Advanced
Bionics,
2012;
Buechner
et
al.
2010).
Study
design
and
procedures
In
this
prospective
efficacy
study,
a
within-subject
repeated
measures
design
was
used.
A
factorial
design
was
defined
with
3
two-level
factors:
factor
TNR
(on/off),
factor
CNR
(on/off),
and
factor
Transients
(stimuli
with
or
without
transients).
A
full
3-factor
design
has
2
3
=
8
conditions,
but
it
was
not
needed
to
test
the
effect
of
factor
TNR
in
combinations
with
stimuli
without
transients
as
the
TNR algorithm
will
not
be
activated
in
these
conditions.
From
the
remaining
six
conditions,
four
conditions
tested
the
different
combinations
of
TNR
and
CNR
for
stimuli
with
transients.
These
four
conditions
were
balanced
across
participants
with
a
4
x
4
Latin
Square.
The
other
two
conditions
tested
CNR-on
and
CNR-off
for
stimuli
without
transients
and
TNR
off.
These
two
conditions
were
alternated
in order
across
participants.
For
all
six
conditions,
the
ANL
and
the
speech
intelligibility
in
noise
were
measured.
After
these
tests
an
annoyance
rating
and
a
paired-comparison
rating
approach
was
used
to
measure
the
effect
of
TNR
and
CNR
on
the
perceived
annoyance
of
four
sounds
that
contained
both
continuous
noise
and
transients.
The
fitting
parameters
of
the
CI
were
set
according
to
the
programme
used
in
daily
life.
If
the
CNR
was
switched
on,
M-levels
were
increased
by
5%
(M-levels
are
basic
fitting
parameters
used
to
define
the
amount
of
electrical
output
at
the
most
comfortable
level).
The
increase
of
M-levels
was
done
in
order
to
increase
the
effect
of
the
CNR,
according
to
the
recommendations
of
Advanced
Bionics
and
previous
research
(Brendel
et
al.
2012;
Dingemanse
and
Goedegebure
2017).
Stimuli
To
test
the
effect
of
TNR,
we
decided
to
use
non-artificial
stimuli
with
pronounced
transients.
A
variety
of
transient
kitchen
sounds
were
recorded
near
a
person's
ear
during
emptying
the
dishwasher
in
a
typical
home
kitchen.
Transients
as
clinking
bowls,
dishes,
cups,
spoons
and
other
similar
sounds
were
recorded
with
a
sample
frequency
of
44.1
kHz
and
a
bit
depth
of
16
bits.
Since
this
was
an
efficacy
study
we
wanted
to
ensure
that
the
TNR
was
activated
by
the
transients.
An
analysis
of
the
fast
envelope
levels
of
the
speech
that
was
used
in
de
speech
intelligibility
and
ANL
tests
showed
that
transients
should
have
a
peak
level
of
at
least
22
dB
above
the
Root
Mean
Square
(RMS)
level
of
the
speech
in
order
to
be
detected
by
the
TNR
algorithm
in
at
least
90%
of
the
cases.
The
RMS-level
of
speech
was
70
dB
(SPL),
so
the
peak
level
of
the
transients
needed
to
be
at
least
92
dB
(SPL).
Transients
that
had
a
lower
peak
level
were
amplified
to
achieve
a
peak
level
of
at
least
92
dB
(SPL).
Transients
that
sounded
unnatural
after
amplification
were
excluded.
Next
it
was
checked
for
which
transients
the
TNR
was
really
activated,
using
the
transients
combined
with
the
speech
signal
of
the
ANL-test
(see
below)
as
input.
This
was
done
by
Advanced
Bionics
with
a
software
implementation
of
the
algorithm.
Eighty-one
per
cent
of
the
transients
activated
the
TNR.
In
other
cases
most
likely
the
rise
time
of
the
transient
was
too
slow
to
reach
the
criterion
of
20
dB/ms.
Again,
these
transients
were
excluded.
At
the
end
of
the
procedure,
there
were
96
unique
transients,
varying
in
content,
duration,
level,
frequency
spectrum
and
experienced
loudness
(see
Table
1
for
details
about
levels).
Note
that
the
transients
were
not
necessary
experienced
as
loud,
because
most
transients
had
a
short
duration.
The
resulting
transient
sounds
were
mixed
with
the
speech
stimuli
for
use
in
the
speech
intelligibility
test
and
the
ANL
test
(see
test
descriptions
for
details).
For
the
paired
comparisons
and
annoyance
ratings,
four
stimuli
were
created
that
were
combinations
of
transients
with
high
peak
levels
and
continuous
noise.
These
stimuli
differed
in
transient
4
J.
G.
Dingemanse
et
al.
Table
1.
Stimuli
and
mean
values
of
the
characteristics
of
the
transients
used.
Type
of
stimuli
Type
of
transients
SPL
peak
(dB
SPL)
#
Transients/
Rise
second
time
(us)
Duration
(ms)
Type
of
continuous
noise
Transients
for
speech-in-noise
tests
Clinking
dishes,
glasses
etc.
95
1.0
675
134
Steady-state
speech
noise
Transients
for
ANL
tests
Clinking
dishes,
glasses
etc.
95
0.5
675
134
Steady-state
speech
noise
Kitchen
sounds
and
exhaust
noise
Clinking
dishes,
glasses
etc.
97
1.6
933
147
Noise
of
an
exhaust
(65
dB
SPL)
Hail
on
car
window
and
car
noise
Hail
hits
on
car
window
97
5.2
437
38
Car
noise
(72
dB
SPL)
Hammering
and
machine
noise
Hammering
96
2.0
1200
195
Noise
of
a
sewing
machine
(63
dB
SPL)
Steps
with
heels
and
babble
Steps
with
heels
97
2.0
1177
198
Babble
noise
100p,
near
continuous
(67
dB
SPL)
Duration
is
the
time
interval
between
the
occurrence
of
transient
peak
level
and
a
level
20
dB
below
this
peak
level.
characteristics
and
in
continuous
noise
type
and
were
thought
to
be
representative
for
different
acoustic
situations
in
daily
life.
Table
1
gives
a
description
of
the
type
and
acoustic
characteristics
of
the
transients
and
continuous
noise.
The
transients
and
the
continuous
sounds
were
mixed
to
create
a
stimulus
in
which
the
transients
were
at
least
22
dB
above
the
continuous
noise
level
in
order
to
be
detected
by
the
TNR
algorithm.
Again,
transients
were
selected
from
recordings
without
additional
signal
processing,
except
some
minor
gain
corrections
to
make
sure
that
transients
were
above
the
threshold
of
the
TNR
activation.
The
four
signals
had
a
duration
of
5
s
and
the
dB
(RMS)
level
was
70
dB
(SPL).
Speech-in-noise
test
Speech
intelligibility
in
noise
was
measured
with
Dutch
female-
spoken,
unrelated
sentences
in
steady-state
speech
spectrum
noise
(Versfeld
et
al.
2000).
The
noise
started
three
seconds
before
the
speech
to
activate
the
CNR
and
ended
0.5
s
after
the
speech.
For
the
speech-in-noise
conditions
with
transients,
a
modified
version
of
the
speech
tracks
was
made
by
applying
four
transients
to
each
list
item.
For
each
list
item
the
four
transients
were
randomly
selected
from
the
set
of
96
transients
(see
previous
paragraph).
Two
of
the
four
transients
were
added
in
the
3-s
interval
of
noise
before
the
start
of
the
sentence,
with
a
randomly
chosen
delay
with
the
constraint
that
the
first
transient
was
within
the
first
half
of
the
interval
and
the
second
transient
in
the
second
half.
This
was
done
to
include
the
possibility
that
the
noise
estimation
of
CNR
ClearVoice
was
influenced
by
the
transients.
The
other
two
transients
were
added
in
the
sentence
interval,
also
with
a
randomly
chosen
delay
and
the
constraint
that
the
first
transient
was
within
the
first
half
of
the
sentence
and
the
second
transient
in
the
second
half.
The
peak
levels
of
the
transients
were
at
least
22
dB
above
the
RMS-level
of
the
speech
to
make
sure
that
the
TNR
was
activated.
The
presen-
tation
level
of
the
sentences
was
fixed
at
70
dB
(SPL).
This
speech
level
is
often
reached
in
noisy
situations
(Pearsons,
Bennett,
and
Fidell
1977).
The
Speech
Reception
Threshold
in
noise
without
transients
(SRTn)
was
measured
twice
with
an
adaptive
procedure
targeting
at
50%
of
words
understood
correctly,
using
26
sentences.
The
first
measurement
was
a
practice
run.
For
the
six
different
test
conditions
in
the
experiment,
the
speech
and
noise
had
a
fixed
SNR
based
on
the
individual
SRTn
+2
dB.
The
2
dB
was
added
because
a
drop
in
intelligibility
due
to
the
transients
was
expected
and
the
test
should
not
be
too
difficult
for
participants.
Furthermore,
floor
and
ceiling
effects
should
be
prevented
for.
Participants
were
asked
to
repeat
as
many
words
as
they
could
from
the
sentence.
The
per
cent
of
correct
words
per
sentence
list
of
18
sentences
was
scored.
Acceptable
noise
level
test
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
ms
of
silence
between
them
and
played
as
running
speech
at
70
dB
SPL
in
all
ANL
measurements.
The
task
was
to
select
the
maximum
background
noise
level
(BNL)
that
the
participant
was
willing
to
accept
while
following
the
speech.
The
listeners
were
given
oral
and
written
instructions,
which
were
Dutch
translations
of
the
instructions
provided
by
Nabelek
et
al.
(2006).
For
each
ANL
measurement
the
BNL
procedure
was
repeated
three
times
and
the
mean
value
was
used
to
calculate
the
ANL
as
the
difference
of
the
speech
level
and
the
mean
BNL.
Before
the
measurements,
participants
were
made
familiar
with
the
BNL
procedure
in
a
practice
condition.
For
the
measurement
conditions
with
transients,
the
transients
were
added
to
the
speech
at
a
rate
of
0.5
Hz.
This
low
rate
was
chosen
to
prevent
the
speech
from
becoming
unintelligible
most
of
the
time,
due
to
the
transients.
The
peak
levels
of
the
transients
were
set
at
least
22
dB
above
the
RMS-level
of
the
speech,
to
make
sure
that
the
TNR
was
activated.
Note
that
the
transient
levels
were
not
changed
in
the
BNL
procedure,
only
the
level
of
the
continuous
noise
was
adjusted,
as
we
wanted
to
be
sure
to
stay
in
the
active
range
of
the
TNR.
Paired
comparisons
and
annoyance
rating
A
paired-comparison
rating
approach
was
used
to
measure
the
effect
of
TNR
and
CNR
on
the
perceived
annoyance
of
four
sounds
that
consisted
of
both
continuous
noise
and
transients.
For
each
sound,
a
participant
compared
three
CI
programmes
with
noise
reduction
(TNR
only,
CNR
only,
TNR
and
CNR
simultaneously)
to
a
reference
condition
without
noise reduction
(TNR-off
and
CNR-
off).
A
two-interval,
seven-alternative
forced
choice
paradigm
was
used,
with
seven
possible
answers
on
an
ordinal
scale,
ranging
from
"A
is
much
less
annoying"
to
"B
is
much
less
annoying".
The
answers
were
transformed
to
numbers
ranging
from
—3
to
3.
The
seven
choice
categories
and
the
transformation
to
numbers
were
in
accordance
with
the
Comparison
Category
Rating
method
described
in ITU-T
P.
800
Annex
E.1
(ITU-T
P.800
1996).
The
participants
could
listen
to
both
fragments
of
sound
as
many
times
as
they
want
C1
C2
C3
•
C4
C5
C6
E<0.001
L<0.001
p
7
b.038
p=0.10
p=0.041
I
,ff0.70
.
E0.67
r
—
—
1
r
— —
1
73,4
68.3
T
100
80
p
ercen
t
correc
t
(mu)
60
40
20
0
Transient
Noise
Reduction
5
CNR
off
CNR
on
CNR
off
CNR
off
CNR
on
CNR
on
TNR
off
TNR
off
TNR
off
TNR
on
TNR
off
TNR
on
no
transients
Figure
1.
Mean
and
95%
confidence
intervals
of
per
cent
correct
scores
for
the
speech
intelligibility
in
noise
test
for
six
conditions.
The
two
light
grey
bars
on
the
left
show
speech
scores
for
speech
without
transients.
The
four
dark
grey
bars
show
speech
scores
values
for
speech
with
transients.
The
annotations
C1-C6
give
the
condition
numbering.
Several
test
conditions
were
compared
and
uncorrected
p-values
were
shown.
Asterisks
denote
that
a
difference
is
significant
after
correction
for
multiple
comparisons.
Dashed
lines
show
the
significance
of
differences
due
to
TNR,
solid
lines
show
the
significance
of
CNR
effects,
and
dash-dotted
lines
show
the
significance
of
the
effect
of
transients.
before
they
completed
their
rating.
They
were
asked
to
listen
to
the
whole
sound
and
to
rate
it
in
the
end.
In
addition,
an
absolute
rating
task
was
used
to
investigate
the
degree
of
annoyance
participants
experienced
in
response
to
the
four
stimuli
used
in
the
paired-comparison
task.
We
asked
the
participants
to
rate
the
experienced
annoyance
on
an
11-point
ordinal
scale.
The
scale
was
labelled
as
"not
at
all
annoying"
at
0,
"slightly
annoying"
at
2.5,
"moderately
annoying"
at
5,
"quite
annoying"
at
7.5,
and
"very
annoying"
at
10,
following
Keidser
et
al.
(2007).
Equipment
Transient
stimuli
were
recorded
with
a
Samson
Q1U
microphone
and
the
audio
editor
Audacity
(2013)
was
used
for
stimulus
preparation.
All
testing
was
performed
in
a
sound-treated
room.
Participants
sat
one
metre
in
front
of
a
Westra
Lab
251
loudspeaker
(Westra
Elektroakustik
GmbH,
Germany)
that
was
connected
to
a
Roland
Octa-capture
soundcard
(model
UA-1010,
Roland
Corporation,
Los
Angeles,
CA),
and
a
computer.
Stimuli
were
presented
in
a
custom
application
(cf.
Dingemanse
and
Goedegebure
2017)
running
in
Matlab
(MathWorks,
v9.0.0).
In
the
ANL
test,
participants
adjusted
the
sound
level
of
the
noise
stimuli
using
the
up
and
down
keys
of
a
keyboard.
The
step
size
for
the
intensity
adjustment
for
the
ANL
task
was
2
dB
per
button
press.
All
participants
were
tested
with
the
same
new
Naida
Q70
processor
and
a
new
T-mic
(Advanced
Bionics,
Stafa,
Switzerland).
Data
analysis
A
priori
power
analysis
using
the
G*Power
software
(Faul
et
al.
2009)
indicated
that
a
sample
of
16
people
would
be
needed
to
detect
a
clinically
significant
ANL
difference
>3
dB
(Olsen
and
Brannstrom
2014)
and
a
clinically
significant
difference
of
10%
points
in
the
word
score
on
a
speech
intelligibility-in-noise
test
with
80%
power
and
alpha
at
0.05.
Speech
performance
scores
were
transformed
to
rationalised
arcsine
unit
(rau)
scores
in
order
to
make
them
suitable
for
statistical
analysis
according
to
(Studebaker
1985).
In
cases
of
multiple
comparisons,
we
used
the
Benjamini—Hochberg
method
to
control
the
false
discovery
rate
at
level
0.05
(Benjamini
and
Hochberg
1995).
Repeated
measures
analysis
of
variance
(RMANOVA)
was
used
to
analyse
the
ANL
and
speech
intelligibility
in
noise
tests.
For
the
analysis
of
the
paired
comparisons
a
one-sample
Wilcoxon
Signed
Rank
test
was
used.
For
the
absolute
annoyance
ratings
a
Friedman
test
was
used
to
detect
if
ratings
were
significantly
different
between
sounds.
Data
interpretation
and
analysis
were
performed
with
SPSS
(IBM,
Version
23,
Chicago,
IL).
Results
Speech
intelligibility
in
noise
A
normality
check
of
the
transformed
per
cent
correct
data
revealed
normally
distributed
data
for
all
conditions.
The
individualised
SNR
ranged
from
2.4
to
18.7
dB.
Figure
1
shows
the
speech
scores
for
the
six
conditions
and
the
significance
levels
of
relevant
differences
between
conditions.
It
is
evident
that
speech
scores
decreased
markedly
with
44%
points
on
average
due
to
the
addition
of
transients.
The
application
of
CNR
lead
to
a
small
increase
in
speech
scores
(6.4%
points
on
average),
but
the
TNR
did
not
alter
I
11.6
12.5
Cl
C2
C3
C4
C5
C6
00.010
.
_
.
_
.E
-
Mga
.
_
.
_
i
p=0.029
.
o=0.002
.
o=0.020
-
pf0.65
r
--
1
CNR
off
TNR
off
I
7.8
CNR
on
TNR
off
15.3
CNR
off
TNR
off
15.9
CNR
off
TNR
on
CNR
on
TNR
off
13.0
CNR
on
TNR
on
30
poor
25
20
15
z
10
good
0
6
J.
G.
Dingemanse
et
al.
no
transients
Figure
2.
Mean
and
95%
confidence
intervals
of
ANL
values.
The
two
light
grey
bars
on
the
left
show
ANL
values
for
speech
without
transients.
The
four
dark
grey
bars
show
ANL
values
for
speech
with
transients.
The
annotations
C1-C6
give
the
condition
numbering.
Several
test
conditions
were
compared
and
uncorrected
p-values
were
shown.
Asterisks
denote
that
a
difference
is
significant
after
correction
for
multiple
comparisons.
Dashed
lines
show
the
significance
of
differences
due
to
TNR,
solid
lines
show
the
significance
of
CNR
effects,
and
dash-dotted
lines
show
the
significance
of
the
effect
of
transients.
the
speech
scores.
A
repeated
measures
ANOVA
with
the
factors
Transients
and
CNR
(conditions
Cl,
C2,
C3,
C5)
showed
a
significant
effect
of
the
Transients
factor
[F(1,15)
=
191.5,
MSE
=
30889.0,
p
<0.001,
77
2
p
=
0.93]
and
a
significant
effect
of
the
CNR
factor
[F(1,15)
=
6.8,
MSE
=
483.1,p
=
0.02,
I/
2
=
0.31].
The
interaction
of
both
factors
was
not
significant
[F(1,15)
=
0.07,
MSE
=
3.6,
p
=
0.80,
77
2
=
0.005].
The
effect
of
TNR,
CNR,
and
the
combined
effect
of
TNR
and
CNR
were
analysed
with
a
second
repeated
measures
ANOVA
with
the
factors
TNR
and
CNR
(conditions
C3, C4,
C5,
C6).
A
significant
effect
was
found
for
the
CNR
factor
[F(1,15)
=
7.8,
MSE
=
805.0,
p=
0.013,
77
2
p
=
0.34],
but
no
significant
effect
was
found
for
the
TNR
factor
[F(1,15)
=
0.003,
MSE
=
0.15,
p
=
0.96,
7/
2
p
<
0.001]
and
the
interaction
of
both
factors
[F(1,15)
=
0.35,
MSE
=
20.2,
p
=
0.57,
I/
2
p
=
0.022].
Acceptable
noise
level
A
normality
check
revealed
that
the
ANL
data
is
normally
distributed
for
each
condition.
Figure
2
presents
the
group
mean
ANL
values
for
the
six
conditions
and
the
significance
levels
of
relevant
differences
between
conditions.
Figure
2
shows
that
in
the
conditions
that
have
transients
added
to
the
speech,
the
noise
tolerance
was
significantly
worsened
compared
to
the
conditions
without
transients
(AANL
=
4.5
dB
on
average).
Switching
on
TNR
did
not
significantly
affect
the
noise
tolerance.
Use
of
the
CNR
significantly
improved
the
ANL
value
with
2.8
dB
on
average
if
transients
were
present
and
3.9
dB
if
transients
were
absent.
A
repeated
measures
ANOVA
with
the
factors
Transients
and
CNR
(conditions
Cl,
C2, C3,
C5)
showed
a
significant
effect
of
the
Transients
factor
[F(1,15)
=
12.0,
MSE
=
318.5,
p=
0.003,
77
2
p
=
0.44]
and
a
significant
effect
of
the
CNR
factor
[F(1,15)
=
15.1,
MSE
=
181.5,
p
=
0.001,
7/
2
p
=
0
.
5
0].
The
interaction
of
both
factors
was
not
significant
[F(1,15)
=
0.93,
MSE
=
5.1,
p=
0.35,
I/
2
p
=
0.059].
The
effect
of
TNR
and
the
combined
effect
of
TNR
and
CNR
(conditions
C3, C4, C5,
C6)
were
analysed
with
a
second
repeated
measures
ANOVA
with
the
factors
TNR
and
CNR.
This
analysis
showed
no
significant
effect
of
the
TNR
factor
[F(1,15)
=
0.49,
MSE
=
2.1,
p
=
0.50,
7/
2
p
=
0.032]
and
a
significant
effect
of
the
CNR
factor
[F(1,15)
=
8.8,
MSE
=
124.2,
p
=
0.010,
?
p
=
0.37]..
The
interaction
of
both
factors
was
not
significant
[F(1,15)
=
0.001,
MSE
=
0.004,
p=
0.98,
7/
2
p
<
0.001].
Substantial
differences
were
found
in
the
noise
tolerance
levels
(ANL-values)
between
the
different
CI-users.
The
reference
ANL
values
(for
CNR-off,
TNR-off
and
no
transients)
ranged
from
5.3
to
20
dB.
No
significant
correlation
was
found
between
the
ANL
(reference
condition
C1)
and
the
median
annoyance
score.
Paired
comparisons
and
annoyance
ratings
Figure
3
shows
the
mean
quantified
rating
score
in
all
three
conditions
for
each
sound
apart
and
for
the
average
over
all
sounds.
Statistical
analysis
was
performed
for
the
ratings
averaged
over
all
the
sounds.
The
programme
with
TNR-on
and
CNR-off
was
rated
as
less
annoying
than
the
reference
condition
(TNR-off;
CNR-off)
for
all
sounds.
This
mean
rating
ranged
between
—1.75
and
0
with
a
median
of
—0.75.
A
Wilcoxon
signed-rank
test
showed
a
statistic-
ally
significant
difference
between
the
median
rating
and
the
test
value
of
0,
z=
—3.3,
p
=
0.001
and
a
large
effect
size
of
r=
—0.8.
The
rating
for
the
TNR-off
CNR-on
programme
ranged
between
—2.25
and
2
with
a
median
of
0.25.
However,
the
Wilcoxon
signed-
rank
test
showed
no
statistically
significant
difference
between
the
median
rating
and
the
test
value
of
0,
z=
1.58,p
=
0.11,
r
=
0.4.
1
1
TNR
on,
CNR
on
TNR
off,
CNR
on
TNR
on,
CNR
off
11
_
Less
annoying
More
_
annoying
1-11-1
Transient
Noise
Reduction
7
mean
of
all
sounds
steps
with
heels
and
babble
hammering
and
machine
noise
hail
on
car
window
and
car
noise
kitchen
sounds
and
exhaust
noise
-3
-2
-1
0
1
2
3
Annoyance
Rating
re
TNR
off
and
CNR
off
Figure
3.
Mean
and
95%
confidence
intervals
of
the
relative
annoyance
rating
scores,
derived
from
the
paired-comparison
data,
for
four
different
sounds.
Each
bar
indicates
the
relative
annoyance
for
a
sound
and
test
condition
compared
with
the
reference
condition
with
TNR-
off
and
CNR-off.
For
the
mean
of
all
sounds,
asterisks
indicate
differences
that
were
significant
on
the
p<
0.05
level.
With
the
combination
of
TNR-on
and
CNR-on
the
annoyance
perception
was
not
different
from
the
reference
condition
on
average,
with
a
median
rating
of
0
and
a
range
from
—1
to
0.75
(Wilcoxon
signed-rank
test,
z=
—0.11,
p
=
0.92,
r=
—0.03).
When
the
three
conditions
were
compared
with
each
other,
the
rating
of
(TNR-off,
CNR-on)
was
significantly
higher
than
the
rating
of
(TNR-on,
CNR-off)
(Wilcoxon
signed-rank
test,
z=
—2.87,
p=
0.002,
r=
—0.5).
The
rating
of
(TNR-on,
CNR-on)
was
also
significantly
higher
than
the
rating
of
(TNR-on,
CNR-off)
(Wilcoxon
signed-rank
test,
z=
—2.86,
p=
0.003,
r
=
—0.5).
The
difference
between
the
rating
of
(TNR-on,
CNR-on)
and
(TNR-off,
CNR-on)
was
nearly
significant
(Wilcoxon
signed-rank
test,
z=
—1.77,
p
=
0.08,
r=
—0.3).
Overall,
the
participants
rated
use
of
TNR
in
the
direction
of
less
annoyance
and
use
of
CNR
in
the
direction
of
more
annoyance.
In
the
absolute
annoyance
rating
task,
the
sounds
were
rated
as
moderately
annoying
on
average.
The
kitchen
sound
was
rated
as
most
annoying
(Median
=
5,
IQR
=
3-6.5),
the
heels
in
babble
as
least
annoying
(Median
=
3.5,
IQR
=
2-6).
The
"hail
on
car
window
and
car
noise"
sound
had
a
median
rating
of
4
(IQR
=
2.5-6)
and
the
"hammering
and
machine
noise"
sound
had
a
median
rating
of
4.5
(IQR
=
3-7).
A
Friedman
test
revealed
a
near
significant
effect
of
type
of
sound
on
annoyance
[X
2
(
3
,
N=
16)
=
6.89,
p
<
0.073].
Ratings
differed
greatly
between
CI
users
with
a
range
from
0
(not
at
all
annoying)
to
10
(very
annoying).
Additionally,
we
analysed
if
higher
annoyance
ratings
were
correlated
with
a
bigger
effect
of
TNR-on
in
the
paired
comparisons
test,
but
no
significant
correlation
was
found.
Discussion
Effects
of
transients
and
need
for
TNR
The
current
study
has
shown
that
transient
sounds
may
be
perceived
as
moderately
annoying
and
substantially
degrade
speech
under-
standing
in
CI
users,
so
there
is
a
need
for
TNR
in
CI-processors.
First,
we
found
an
average
annoyance
rating
in
CI
recipients
for
transient
sounds
of
4.5
(moderate
annoyance)
on
an
11-point
scale,
which
is
lower
than
the
reported
annoyance
scores
of
6.3
for
average
to
loud
transient
sounds
in
new
wearers
of
hearing
aids
(Hernandez,
Chalupper,
and
Powers
2006).
An
explanation
for
this
difference
may
be
that
the
participants
of
this
study
were
experienced
CI
users,
who
were
more
used
to
hearing
average
to
loud
sounds
than
new
wearers
of
hearing
aids.
Furthermore,
the
AGC
of
the
CI-processor
used
had
a
fast
compressor
with
a
compression
ratio
of
12
above
71
dB
SPL,
which
prevents
sounds
becoming
too
loud.
In
hearing
aids,
compression
ratios
are
much
lower
and
consequently
high
input
levels
may
cause
more
annoyance.
Still,
in
CI
users,
TNR
may
be
helpful
to
reduce
the
perceived
level
of
annoyance
of
transient
sounds.
Second,
the
presence
of
high
level
transients
caused
a
large
decrease
in
speech
intelligibility
in
noise.
Activation
of
the
AGC
may
be
the
main
explanation
of
this
result.
The
transients
in
our
experiment
had
durations
that
were
long
enough
to
activate
the
fast
compressor
(attack
time
3
ms).
The
fast
compressor
has
a
release
time
of
80
ms
and
affected
at
least
one
word
in
the
sentences.
Due
to
the
high
transient
peak
levels
and
the
high
"transient-to-speech-
ratio"
of
at
least
22
dB
in
our
experiment,
the
AGC
attenuated
the
speech
level
to
just
below
50
dB
SPL.
At
this
speech
level,
average
speech
intelligibility
in
noise
for
CI
users
is
relatively
low
at
20%,
according
to
Boyle,
Nunn,
and
O'Connor
(2013).
Our
results differ
from
the
findings
of
Stobich,
Zierhofer,
and
Hochmair
(1999)
who
reported
word
scores
between
50
and
60%
for
speech
with
a
transient
and
different
AGC
configurations.
However,
they
used
only
one
transient
at
the
beginning
of
the
sentence,
a
"transient-to-
speech-ratio"
of
15
dB
and
a
compression
ratio
of
3
or
6.
Another
reason
that
may
have
contributed
to
the
drop
in
intelligibility
could
be
the
masking
of
the
speech
signal
by
the
transients.
It
is
likely
that
forward
masking
occurred
besides
simultaneous
masking,
because
the
transient
levels
were
much
louder
than
the
speech
level.
The
recovery
of
masking
in
CI
users
is
8
J.
G.
Dingemanse
et
al.
thought
to
be
a
process
in
the
central
auditory
system
(Dingemanse,
Frijns,
and
Briaire
2006;
Lee,
Friedland,
and
Runge
2012;
Shannon,
1990).
The
time
required
for
recovery
of
masking
is
highly
variable
between
CI
users
and
ranges
between
100
ms
and
more
than
1
s
making
it
likely
that
forward
masking
played
a
role,
at
least
for
some
patients.
The
finding
that
transients
were
highly
disruptive
for
speech
perception
is
clinically
important.
Many
of
the
participants
reported
that
they
experience
a
comparable
disrupting
effect
of
transient
sounds
when
listening
to
speech
in
daily
life.
This
emphasises
the
need
for
an
effective
TNR
algorithm
in
CI
processors
that
is
able
to
(partly)
compensate
for
the
detrimental
effect
of
transients
on
speech.
Third,
the
presence
of
transients
caused
a
moderate
decrease
in
noise
tolerance
(increase
of
ANL).
It
is
most
likely
that
reduced
speech
intelligibility
played
an
important
role
in
the
observed
decrease
in
noise
tolerance.
The
ANL
test
has
an
instruction
that
contains
the
words
"while
following
the
story",
indicating
that
intelligibility
of
the
speech
is
required
in
the
ANL
test.
Although
the
rate
of
transients
was
half
of
that
in
the
speech-in-noise
test,
transients
made
parts
of
the
speech
unintelligible,
which
made
it
more
difficult
to
follow
the
speech.
Therefore,
there
was
less
room
for
adding
noise
that
further
reduces
speech
intelligibility.
In
addition,
the
combination
of
transients
and
noise
may
be
less
tolerable
than
noise alone.
Effects
of
TNR
This
study
has
shown
that
application
of
TNR
can
lead
to
significantly
reduced
perceived
annoyance
for
mixtures
of
natural
transient
sounds
with
high
peak
levels
and
continuous
noises.
This
finding
is
in
accordance
with
the
intended
effect
of
the
algorithm
and
confirms
the
efficacy
of
the
algorithm.
The
amount
of
annoyance
reduction
was
—0.75
on
average
compared
to
the
condition
without
TNR,
which
should
be
interpreted
as
slightly
better,
according
to
the
Comparison
Category Rating
scale
described
in
ITU-T
P.
800
Annex
E.1
(ITU-T
P.800
1996).
This
is
only
a
small
improvement,
but
it
is
relative
to
the
moderate
annoyance
without
TNR.
A
small
improvement
still
can
contribute
to
improved
listening
comfort
in
daily
practice.
Perceived
annoy-
ance
of
transients
substantially
differed
between
individual
CI
users.
This
means
that
some
users
did
not
profit
from
TNR
as
they
hardly
perceived
the
transients
as
annoying,
while
the
other
CI-users
that
do
need
TNR
may
have
profited
substantially.
Although
TNR
was
able
to
reduce
perceived
annoyance,
the
application
of
TNR
had
no
significant
effect
on
noise
tolerance
or
speech
intelligibility
in
noise
in
this
study.
This
is
in
contrast
with
Dyballa
et
al.
(2015)
who
reported
a
small
but
significant
improvement
of
0.4
dB
in
SRTn
for
speech
intelligibility
in
dish-
clinking
transient
noise,
using
a
comparable
TNR
algorithm.
They
used
a
speech
material
that
was
easier
to
recognise,
which
consisted
of
50
words
that
participants
knew
from
training.
Possibly
this
made
their
test
more
sensitive
to
small
changes.
In
agreement
with
the
results
of
this
study,
Keidser
et
al.
(2007)
reported
that
the
TNR
had
no
significant
effect
on
speech
recognition
in
background
noise
in
hearing
aid
users.
Furthermore,
the
lack
of
an
effect
for
noise
tolerance
and
speech
intelligibility
in
noise
in
this
study
may
be
due
to
the
short
duration
of
the
signal
reduction
by
the
TNR
compared
to
the
duration
of
the
transients.
If
a
transient
is
detected,
TNR
attenuates
the
signal
by
14
or
20
dB,
but
within
5
ms
this
attenuation
is
reduced
to
about
5
dB,
because
of
the
short
time
constant
and
the
exponential
reduction
of
the
TNR
attenuation.
Therefore,
the
effect
of
the
AGC
and
the
amount
of
masking
would
be
largely
the
same
for
the
TNR-on
and
TNR-off
conditions.
An
improvement
in
the
TNR
algorithm
could
be
made
so
that
the
attenuation
reduction
follows
the
decrease
in
level
of
the
fast
signal
envelope
that
is
used
in
the
algorithm.
This
may
prevent
activation
of
the
AGC,
which
has
a
longer
release
time
than
the
TNR
algorithm.
As
a
result,
transients
may
be
less
detrimental
for
speech
intelligibility.
Using
a
shorter
release
time
of
the
AGC
could
be
another
option
to
reduce
the
detrimental effect
of
transients
on
speech
perception.
Interaction
of
TNR
and
CNR
The
combined
application
of
TNR
and
CNR
did
not
result
in
a
cumulated
improvement
of
speech
intelligibility
in
noise
for
CI-
users.
This
is
in
accordance
with
the
absence
of
an
effect
of
TNR
alone.
Furthermore,
the
effect
of
CNR
was
not
influenced
by
the
application
of
TNR.
An
possible
explanation
for
this
finding
is
that
on
the
moment
of
a
transient,
speech
intelligibility
is
disturbed,
regardless
of
the
effect
of
TNR
on
the
CNR.
In
the
paired-comparison
experiment,
participants
perceived
more
annoyance
on
average
(although
not
significant)
with
CNR
on
compared
with
the
reference
condition
(CNR-off,
TNR-off)
in
noisy
backgrounds
that
contained
transient
sounds.
This
is
most
likely
due
to
an
increase
in
M-levels
of
5%
in
the
CNR-on
programmes.
The
combined
application
of
TNR
and
CNR
resulted
in
an
equal
annoyance
perception
for
the
conditions
(TNR-on,
CNR-on)
and
(TNR-off,
CNR-off),
indicating
that
the
increased
annoyance
that
arose
from
the
increased
M-levels
was
compensated
for
by
the
use
of
TNR.
This
shows
that
TNR
may
be
helpful
in
combination
with
CNR,
as
it
prevents
CI-users
from
substantially
turning
down
the
volume
due
to
annoyance
to
transient
sounds.
These
findings
suggest
to
apply
CNR
and
TNR
together
with
a
5%
M-level
increase
in
a
clinical
used
speech
in
noise
programme,
to
optimise
both
speech
understanding
and
listening
comfort
in
noise.
General
discussion
and
conclusions
This
study
was
designed
as
an
efficacy
study
to
investigate
the
effect
of
a
TNR algorithm
and
its
necessity
by
investigating
the
annoyance
and
detrimental
effect
of
the
transients
that
were
reduced
by
the
TNR.
The
large
disturbing
effect
of
transients
on
speech
intelligibility
in
noise
and
the
positive
effect
of
TNR
on
noise
annoyance
we
found
in
our
study
shows
that
it
is
worthwhile
to
further
study
the
perception
of
transient
sounds
and
effects
of
TNR
in
CI
users.
A
limitation
of
this
study
is
that
only
transients
with
high
peak
levels
were
used.
This
is
only
a
subset
of
transients
that
occur
in
daily
life.
It
is
expected
that
transients
with
lower
peak
levels
are
less
annoying
and
less
detrimental
for
speech
perception.
Future
studies
should
investigate
the
effect
of
transients
on
speech
in
quiet
and
noise
at
several
speech
levels
and
several
"transient-to-
speech"
ratios
to
get
more
insight
in
the
detrimental
effects
of
transients
on
speech
perception
in
CI
users.
They
should
also
investigate
more
in
general
how
transients
are
perceived
by
CI-
users,
and
what
factors
may
improve
the
listening
conditions
in
the
presence
of
transients.
Furthermore,
it
should
be
noted
that
CI
users
may
prefer
to
perceive
some
transients,
like
transients
in
music
or
in
alarm
signals.
Also,
transients
may
be
important
cues
in
sound
perception
and
TNR
should
not
disrupt
these
cues.
Ultimately,
field
studies
should
be
used,
investigating
both
disrupting
and
positive
Transient
Noise
Reduction
9
effects
of
transients
and
possible
improvements
or
negative
side
effects
of
TNR.
Smart
algorithms
based
on
sound
environment
classification
would
be
a
desirable
development.
Another
limitation
of
this
study
is
that
we
included
good
performers
only
(CVC
scores
>70%).
The
effect
of
the
CNR
and
TNR
algorithms
is
not
necessarily
the
same
for
CI
users
with
less
benefit
of
the
CI.
These
CI
users
complain
more
often
that
sounds
are
too
loud
or
too
disturbing,
so
there
is
more
room
for
improvement,
at
least
for
listening
comfort.
On
the
other
hand,
the
effect
of
TNR
may
be
too
small
to
really
cause
a
significant
shift
in
listening
comfort
and
performance
as
noisy
conditions
remain
extremely
challenging
for
this
group
of
CI
users
We
conclude
that
the
investigated
TNR
algorithm
in
a
CI
processor
was
effective
in
reducing
annoyance
from
transient
sounds
with
high
peak
levels,
without
causing
a
negative
effect
on
speech
understanding.
However,
TNR
was
not
able
to
compensate
for
the
large
decrease
in
speech
understanding
caused
by
transient
sounds.
TNR
did
not
reduce
the
beneficial
effect
of
CNR
on
speech
intelligibility
in
noise,
but
no
cumulated
improvement
was
found
either.
Both
types
of
noise
reduction serve
different
goals
and
work
independently,
so
they
can
be
easily
combined
in
one
CI
system.
Acknowledgements
The
authors
gratefully
acknowledge
the
participants
and
they
thank
Phillipp
Hehrmann
for
analysis
of
the
signals
with
respect
of
TNR
activation
and
Martina
Brendel
for
her
helpful
comments.
Portions
of
the
data
were
presented
at
the
14th
International
Conference
on
Cochlear
Implants
(CI2016),
May
11-14,
2016,
Toronto.
Declaration
of
interest:
This
work
was
supported
by
Advanced
Bionics.
ORCID
J.
Gertjan
Dingemanse
e
http://orcid.org/0000-0001-8837-3474
References
Advanced
Bionics.
2012.
ClearVoice.
Technical
Facts.
Valencia,
CA:
Advanced
Bionics.
Audacity.
2013.
Audacity
(Version
2.0.3)
[Computer
Program].
http://
audacity.sourceforge.net/
[Downloaded
4
March
2013].
Benjamini,
Y.,
and
Y.
Hochberg.
1995.
"Controlling
the
False
Discovery
Rate:
A
Practical
and
Powerful
Approach
to
Multiple
Testing."
Journal
of
the
Royal
Statistical
Society.
Series
B
(Methodological)
57
(1):
289-300.
doi:10.2307/2346101.
Bosman,
A.
J.,
and
G.
F.
Smoorenburg.
1995.
"Intelligibility
of
Dutch
CVC
Syllables
and
Sentences
for
Listeners
with
Normal
Hearing
and
with
Three
Types
of
Hearing
Impairment."
International
Journal
of
Audiology
34
(5):
260-284.
doi:10.3109/00206099509071918.
Boyle,
P.
J.,
A.
Buchner,
M.
A.
Stone,
T.
Lenarz,
and
B.
C.
J.
Moore.
2009.
"Comparison
of
Dual-Time-Constant
and
Fast-Acting
Automatic
Gain
Control
(AGC)
Systems
in
Cochlear
Implants."
International
Journal
of
Audiology
48
(4):
211-221.
doi:10.1080/14992020802581982.
Boyle,
P.
J.,
T.
B.
Nunn,
A.
F.
O'Connor,
and
B.
C.
Moore.
2013.
"STARR:
A
Speech
Test
for
Evaluation
of
the
Effectiveness
of
Auditory
Prostheses
Under
Realistic
Conditions."
Ear
&
Hearing
34
(2):
203-
212.
doi:10.1097/AUD.0b013e31826a8e82.
Brendel,
M.,
T.
Rottmann,
C.
Frohne-Buechner,
A.
Buechner,
and
T.
Lenarz.
2012.
"Benefit
of
the
Noise
Reduction
Algorithm
ClearVoice
for
Experienced
Cochlear
Implant
Users."
Paper
Presented
at
the
12th
International
Conference
on
Cochlear
Implants
and
Other
Implantable
Auditory
Technologies,
Baltimore,
USA.
Buechner,
A.,
M.
Brendel,
H.
Saalfeld,
L.
Litvak,
C.
Frohne-Buechner,
and
T.
Lenarz.
2010.
"Results
of
a
Pilot
Study
with
a
Signal
Enhancement
Algorithm
for
HiRes
120
Cochlear
Implant
Users."
Otology
&
Neurotology
31
(9):
1386-1390.
DiGiovanni,
J.
J.,
E.
A.
Davlin,
and
N.
K.
Nagaraj.
2011.
"Effects
of
Transient
Noise
Reduction
Algorithms
on
Speech
Intelligibility
and
Ratings
of
Hearing
Aid
Users."
American
Journal
of
Audiology
20
(2):
140-150.
doi:10.1044/1059-0889(2011/10-0007).
Dingemanse,
J.
G.,
J.
H.
Frijns,
and
J. J.
Briaire.
2006.
"Psychophysical
Assessment
of
Spatial
Spread
of
Excitation
in
Electrical
Hearing
with
Single
and
Dual
Electrode
Contact
Maskers."
Ear
and
Hearing
27
(6):
645-657.
doi:10.1097/01.aud.0000246683.29611.1b.
Dingemanse,
J.
G.,
and
A.
Goedegebure.
2017.
"Optimising
the
Effect
of
Noise
Reduction
Algorithm
ClearVoice
in
Cochlear
Implant
Users
by
Increasing
the
Maximum
Comfort
Levels."
International
Journal
of
Audiology.
Advance
online
publication.
doi:10.1080/14992027.2017.
1390267.
Dyballa,
K.-H.,
P.
Hehnnann,
V.
Hamacher,
W.
Nogueira,
T.
Lenarz,
and
A.
Buchner.
2015.
"Evaluation
of
a
Transient
Noise
Reduction
Algorithm
in
Cochlear
Implant
Users."
Audiology
Research
5
(2):
116.
doi:10.408
1
/
a
udiores.2015.116.
Faul,
F.,
E.
Erdfelder,
A.
Buchner,
and
A.-G.
Lang.
2009.
"Statistical
Power
Analyses
Using
G*
Power
3.1:
Tests
for
Correlation
and
Regression
Analyses."
Behavior
Research
Methods
41
(4):
1149-1160.
doi:10.3758/BRM.41.4.1149.
Hernandez,
A.,
J.
Chalupper,
and
T.
Powers.
2006.
"An
Assessment
of
Everyday
Noises
and
Their
Annoyance."
Hearing
Review
13,
16.
ITU-T
P.800.
1996.
Methods
for
Subjective
Determination
of
Transmission
Quality.
International
Telecommunications
Union,
Telecommunications
Standardization
Sector,
Annex
E.1,
p.
24.
Keidser,
G.,
A.
O'Brien,
M.
Latzel,
and
E.
Convery.
2007.
"Evaluation
of
a
Noise-Reduction
Algorithm
that
Targets
Non-Speech
Transient
Sounds."
The
Hearing
Journal
60
(2):
29-32.
doi:10.1097/
011110000285643.45157.35.
Khing,
P.
P.,
B.
A.
Swanson,
and
E.
Ambikairajah.
2013.
"The
Effect
of
Automatic
Gain
Control
Structure
and
Release
Time
on
Cochlear
Implant
Speech
Intelligibility."
PLoS
One
8
(11):
e82263.
doi:10.1371/
journal.pone.0082263.
Korhonen,
P.,
F.
Kuk,
C.
Lau,
D.
Keenan,
J.
Schumacher,
and
J.
Nielsen.
2013.
"Effects
of
a
Transient
Noise
Reduction
Algorithm
on
Speech
Understanding,
Subjective
Preference,
and
Preferred
Gain."
Journal
of
the
American
Academy
of
Audiology
24
(9):
845-858.
doi:10.3766/
jaaa.24.9.8.
Lee,
E.
R.,
D.
R.
Friedland,
and
C.
L.
Runge.
2012.
"Recovery
from
Forward
Masking
in
Elderly
Cochlear
Implant
Users."
Otology
&
Neurotology
33
(3),
355-363.
doi:10.1097/MA0.0b013e318248ede5.
Liu,
H.,
H.
Zhang,
R.
A.
Bentler,
D.
Han,
and
L.
Zhang.
2012.
Evaluation
of
a
Transient
Noise
Reduction
Strategy
for
Hearing
Aids.
Journal
of
the
American
Academy
of
Audiology
23
(8):
606-615.
doi:10.3766/
jaaa.23.8.4.
Mauger,
S.
J.,
K.
Arora,
and
P.
W.
Dawson.
2012.
"Cochlear
Implant
Optimized
Noise
Reduction."
Journal
of
Neural
Engineering
9
(6):
065007.
doi:10.1088/1741-2560/9/6/065007.
Mertens,
G.,
A.
Kleine
Punte,
M.
De
Bodt,
and
P.
Van
de
Heyning.
2015.
"Sound
Quality
in
Adult
Cochlear
Implant
Recipients
Using
the
HISQUI19."
Acta
Oto-Laryngologica
135
(10):
1138-1145.
doi:
10.3109/00016489.2015.1066934.
Moore,
B.
C.,
B.
R.
Glasberg,
and
M.
A.
Stone.
1991.
"Optimization
of
a
Slow-Acting
Automatic
Gain
Control
System
for
Use
in
Hearing
Aids."
British
Journal
of
Audiology
25
(3):
171-182.
doi:10.3109/
03005369109079851.
Nabelek,
A.
K.,
M.
C.
Freyaldenhoven,
J.
W.
Tampas,
S.
B.
Burchfiel,
and
R.
A.
Muenchen.
2006.
"Acceptable
Noise
Level
as
a
Predictor
of
10
J.
G.
Dingemanse
et
al.
Hearing
Aid
Use."
Journal
of
the
American
Academy
of
Audiology
17
(9):
626-639.
doi:10.3766/jaaa.17.9.2.
Olsen,
S.
0.,
and
K.
J.
Brarmstrom.
2014.
"Does
the
Acceptable
Noise
Level
(ANL)
Predict
Hearing-Aid
Use?"
International
Journal
of
Audiology
53
(1):
2-20.
doi:10.3109/14992027.2013.839887.
Pearsons,
K.
S.,
R.
L.
Bennett,
and
S.
Fidell.
1977.
Speech
Levels
in
Various
Noise
Environments.
Washington,
DC:
U.S.
Environmental
Protection
Agency.
Shannon,
R.
V.
1990.
"Forward
Masking
in
Patients
with
Cochlear
Implants."
The
Journal
of
the
Acoustical
Society
of
America
88
(2):
741-744.
doi:10.1121/1.399777.
Stobich,
B.,
C.
M.
Zierhofer,
and
E.
S.
Hochmair.
1999.
"Influence
of
Automatic
Gain
Control
Parameter
Settings
on
Speech
Understanding
of
Cochlear
Implant
Users
Employing
the
Continuous
Interleaved
Sampling
Strategy."
Ear
and
Hearing
20
(2):
104-116.
doi:10.1097/
00003446-199904000-00002.
Stone,
M.A.,
B.
C.
J.
Moore,
J.
I.
Alcantara,
and
B.
R.
Glasberg.
1999.
"Comparison
of
Different
Forms
of
Compression
Using
Wearable
Digital
Hearing
Aids."
The
Journal
of
the
Acoustical
Society
of
America
106
(6):
3603-3619.
doi:10.1121/1.428213.
Studebaker,
G.
A.
1985.
"A
"Rationalized"
Arcsine
Transform."
Journal
of
Speech
and
Hearing
Research
28
(3):
455-462.
doi:10.1044/
j
shr.2803
.455.
Vaerenberg,
B.,
P.
J.
Govaerts,
T.
Stainsby,
P.
Nopp,
A.
Gault,
and
D.
Gnansia.
2014.
"A
Uniform
Graphical
Representation
of
Intensity
Coding
in
Current-Generation
Cochlear
Implant
Systems."
Ear
and
Hearing
35
(5):
533-543.
doi:10.1097/AUD.0000000000000039.
Versfeld,
N.
J.,
L.
Daalder,
J.
M.
Festen,
and
T.
Houtgast.
2000.
"Method
for
the
Selection
of
Sentence
Materials
for
Efficient
Measurement
of
the
Speech
Reception
Threshold."
The
Journal
of
the
Acoustical
Society
of
America
107
(3):
1671-1684.
doi:10.1121/
1.428451.