Download What is Frequency Lowering? - Indiana Speech-Language

Overview: Frequency Lowering
Frequency Lowering
in Hearing Aids
Joshua M. Alexander, Ph.D., CCC-A
2012 ISHA Convention
What is it?
Why use it?
Techniques in hearing aids
Research findings
Verification
March 29-31
Indianapolis, IN
www.TinyURL.com/PurdueEAR
What is Frequency Lowering?
Concept Behind Frequency Lowering
Amplified Speech Spectrum
A method of increasing access to a wider range of
spectral content
Picks up where traditional amplification leaves off
Level, dB SPL
Amplification ‘moves’ the level of soft sounds into a
range that is audible for the patient
Frequency lowering moves the spectral content of
high-frequency sounds to a region that is audible for
the patient
Inaudible
Average Speech
Frequency, Hz
Research Interest
Frequency Lowering in Hearing Aids
1990s
Experimental
Vocoding
1980s
and slow playback techniques (1960-1970’s)
1970s
1960s
Wearable device
Oticon
2000-2009
Not a new concept for severe hearing loss
0
TP 72 (Johannson, 1961)
Oldest technique currently in use: since 1991
5
10
15
20
# of Articles on Frequency Lowering
25
Source:
Lejon (2009)
Why the renewed interest?
New technologies allow for new techniques
New research findings
New target population: mild-moderate hearing loss
1
Industry Interest
Controversy
# of Hearing Aid Models with Frequency Lowering
20
18
While the speech code is relatively ‘scale
invariant,’ it is heavily dependent on frequency
No currently implemented DSP strategy has as much
potential to change the identity of individual speech
sounds
Potential to make speech understanding worse because
low-frequency information has to be altered to
accommodate displaced high-frequency information
16
14
12
10
8
6
Widex
June 2008
First products for mild
to moderate losses
4
2
0
1991-1998 2000-2006
2006
2007
2008
2009
2010
2011
Why Even Consider It?
Rationale varies depending on severity of loss
Moderately severe or greater loss
Precipitous high-frequency loss
Mild to moderate loss
Potential Benefits of
Frequency Lowering
What do we hope to achieve?
Influenced by the information we hope to
‘recover’ from the speech spectrum
Answer: some of the same benefits as
increasing audible bandwidth with traditional
amplification- if we could
Different goals for different hearing losses
Moderate to Severe Loss
Moderate to Severe Loss
Rationale: make some of this information available to the user
5k
Inaudible
4k
Level, dB SPL
Inaudible
Average Speech
3k
2k
1k
ch
i l d
r
en
l i ke s t r aw b err ie
/z/
Frequency, Hz
2
High-Frequency Dead regions
Moderately Severe Losses (or worse)
Limitations in hearing aid gain and feedback
Important mid-frequency information is inaudible
High-frequency dead regions
May not make as effective use of amplified highfrequency information and may even perform worse
when made audible
E.g.,
A form of uncontrolled frequency lowering (transposition)
Signal detected at an off-frequency place on the basilar membrane
Detection occurs at same place where information is processed
more normally
Unknown re-coding process: tonal stimuli may sound noiselike/distorted
Moore, 2004; Baer et al., 2002; Vickers et al., 2001
Signal Freq.
Rationale
Provide users with information that is normally
available to those with less severe hearing loss (~5kHz)
Avoid dead regions
(Kluk & Moore, 2006)
Precipitous Losses
Precipitous Losses
-10
Need help that conventional amplification cannot
provide
0
10
20
Too Normal
Hearing Level, dB HL
30
40
Rationale: increase
“information value” of
what’s being amplified
50
60
Marginal candidates for amplification
Limited frequency range with which to amplify speech
because of difficulty getting sufficient gain in regions of
hearing loss
70
Too normal to amplify and too severe to amplify
80
90
100
Too Severe
110
120
Rationale
250
500
1000
2000
4000
8000
Frequency, Hertz
Help users make better use of a very limited residual
bandwidth of audibility (e.g., < 2000 Hz)
Mild to Moderate Losses
A novel application of frequency lowering
Extend bandwidth of information to ‘ultra’ high
frequencies beyond that normally achievable with
conventional amplification (>5 kHz)
Audible bandwidth limited by hearing aid, not by
hearing loss per se:
Receiver output
Maximum gain without feedback
Manufacturer decisions to limit DSP (sampling rate, processing
delay, current drain, etc.)
Hearing Aid Limitations
Moore et al. (2008): To partially restore audibility
for 65 dB hearing loss at 10 kHz requires an effective
insertion gain of 36 dB (90 dB SPL)
Could only be achieved in 25 of 62 ears
68 dB HL average at 10 kHz
Possible unpleasant, harsh, or tinny sound quality, or
loudness discomfort
Wireless streaming limitations
Audio bandwidth of Bluetooth is 7.5 kHz
3
Typical HA Receiver Response
Before Filtering
+6 dB/oct. high-frequency emphasis
10k
0
9k
-5
Relative Level, dB SPL
8k
-10
7k
6k
-15
5k
-20
4k
Gain is least where
hearing loss is greatest
-25
3k
2k
-30
1k
-35
250
500
1000
2500
5000
10,000
Frequency, Hz
ch
i l d
Filtered by Receiver Response
9k
0.3
0.25
Band Importance
Inaudible
8k
7k
6k
5k
Rationale: make
frication audible
4k
en
l i ke s t r aw b err ie
/z/
What’s Being Missed?
+6 dB/oct. high-frequency emphasis
10k
r
Nonsense Syllables
Short Passages
0.2
0.15
~13%
0.1
SII/AI: ANSI
S3.5-1997
(R2007)
0.05
3k
2k
0
<225
1k
225-565
565-1125
1125-2250
Frequency Bands, Hz
2250-4500
>4500
A decrease in redundancy and context increases the relative
importance of high-frequency information
Results are for filtered speech in quiet; effects of noise?
ch
i l d
r
en
l i ke s t r aw b err ie
/z/
What’s Being Missed?
Voiceless Fricative Peak Frequencies
Bandwidth of /s/ at
Max Performance
Importance of High Frequencies
Even
NH adults listening in quiet benefit from high
frequency information
Relative importance increases in challenging listening
conditions and with decreased linguistic context
and affricates account for about 50% of English
consonants
Fricatives
/s/,
/z/: about 8% of spoken English consonants (Rudmin, 1983)
implications for morphological development
Significant
Fox & Nissen (2005)
Stelmachowicz et al. (2001)
Over 20 linguistic uses for /s/, /z/: plurality, third person present
tense, past vs. present tense, to show possession, possessive
pronouns, contractions (Rudmin, 1983)
4
Developmental Importance
Developmental Importance
Fricative Production: Despite early intervention,
toddlers with mild to moderate hearing loss
produce very few fricatives or affricates (Moeller
et al., 2007)
Moeller et al., 2007
While they start to catch up with their normalhearing peers on production of other sound classes,
they fall further behind in production of fricatives
and affricates
Developmental Importance
Morpho-Syntactic Development: 5 to 7 yearolds identified late (2-3 years) less consistently
and less accurately use the morpheme /s/
(Moeller et al., 2010)
Production accuracy of /s/, /z/ in children with
hearing aids is 15-18% worse than children with
CI’s (Grant et al., 2002)
Developmental Importance
Conclusion:
Conventional hearing aids do not provide
the high-frequency cues necessary
for normal development of speech
even cases of mild to moderate loss
Benefit of Extending Bandwidth
Benefit of Extending Bandwidth
Increasing BW >5 kHz in the laboratory (no lowering)
Improved speech recognition
Increasing BW >5 kHz in the laboratory (no lowering)
Improved speech clarity and music quality
Especially fricatives (Stelmachowicz & colleagues)
Sentences in quiet/noise (Hornsby & Ricketts,
2006/Hornsby, 2007)
May depend on audiometric configuration (Hornsby et
al., 2011)
Improved novel word learning by children (Pittman
2008)
Moore
& Tan (2003):
For normals, upper cutoff for speech =
16.5 kHz
Ricketts
10.8 kHz and for music =
et al. (2008)
Impaired listeners preferred 9-kHz bandwidth when slope of loss
was <7.5 dB/oct.
Sjolander
& Holmberg (2009)
Listeners using a hearing aid with extended high-frequency gain
generally preferred a 10-kHz bandwidth over 8-kHz bandwidth
for listening to music
Füllgrabe
et al. (2010); Moore et al. (2011)
5
Benefit of Extending Bandwidth
Benefit of Extending Bandwidth
Increasing BW >5 kHz in the laboratory (no lowering)
Less effortful listening (Karlsen et al., 2006)
It remains to be seen whether
these same benefits be achieved
using frequency lowering
Improved localization (Best et al., 2005; Brungart &
Simpson, 2009)
Pinna cues: elevation, front/back
Improved spatial unmasking (Hamacher et al.,
2005; Moore et al., 2010)
Head shadow increases with frequency
Mild to Moderate Losses
Rationale for frequency lowering
Provide users with information that is beyond the
bandwidth normally achievable with conventional
amplification (>5 kHz)
Summary of What and Why
Improve perception (and production) of fricatives,
especially /s/
Possibly reduce listening fatigue and improve sound
quality by providing a fuller, richer signal
Make better use of a very restricted bandwidth of audibility
Moderately severe or greater losses
Hearing loss and other reductions in redundant speech information and
context (e.g., noise and maturation) increase the importance and need
for high-frequency information
Precipitous losses
Why use frequency lowering?
Extend the bandwidth of information to that which is normally
achievable with conventional amplification (~5 kHz)
Avoid excessively amplifying dead regions
Mild to moderate losses
Extend the bandwidth of information beyond the bandwidth normally
achievable with conventional amplification (>5 kHz)
Improve perception (production) of fricatives, especially /s/, which are
a significant problem, despite early intervention
Terminology
Source Region
The frequency range that is subject to lowering
Techniques for Frequency
Lowering in Hearing Aids
The frequency range where lowered information is moved to
Frequency Compression (squeeze)
Start frequency: lowest frequency in the source region
Target Region
Target region is contained within the source region
Bandwidth of the source region is reduced
Start frequency (even 0 Hz) is like an anchor that does not move
Frequency Transposition/Spectral Feature Translation
Less overlap between target and source regions
Bandwidth of the source region is not reduced
Start frequency is moved to a lower frequency
(copy & paste)
6
Techniques in Today’s Hearing Aids
Techniques in Today’s Hearing Aids
Dynamic Linear Frequency Compression (DLFC)
AVR Sonovation: “Dynamic Speech Recoding (DSR),” or “Digital
Frequency Compression (DFC),” or “Frequency Compression for
Voicelessness” (FCV/FCVL) - Introduced in 1991 (body), 1998/2000
(BTE/ITE), 2004 (DSP)
Inaudible
Time
Nonlinear
Frequency
Compression
Spectral
Feature
Translation
Time
Time
Frequency Transposition (FT)
Widex: “Audibility Extender” (AE) - Introduced in 2006
Frequency
Frequency
Linear
Frequency
Transposition
Frequency
Frequency
Dynamic Linear
Frequency
Compression
Nonlinear Frequency Compression (NFC)
Phonak: “Sound Recover” (SR) - Introduced in 2008
Time
Spectral Feature Translation
Starkey: “Spectral iQ” - Introduced in 2011
A Classification Scheme
Dynamic Linear Freq. Compression
Technique
Activation
Input Dependent
Always Active
Compression
AVR
(linear @ 0 Hz)
Phonak
(nonlinear @ start)
Transposition
Starkey
(feature lowering)
Widex
(peak lowering)
Proportional on a linear scale
‘Frequency divider’: e.g., DFC = 5.0 means that 5 kHz will
move to 1 kHz (or, 8 kHz to 1.6 kHz) pitch is changed
Spectral peak relationships are maintained within a
compressed segment
Over-sampled input leads to “slow-play” effect
Sampling rate output < input by (1/DFC)
From 0 Hz to upper limits of mic & DSP
Output
Samples are played
out at a slower rate
Input
All when energy >2.5 kHz is greater than <2.5 kHz, otherwise none
Most likely noisy fricatives and bursts associated with stop and affricate
consonants
Dominant low frequency energy (especially vowels) is unaltered more
natural sound quality
Pitch for voiced speech is mostly unaltered
“Dynamic consonant boost” to increase lowered-signal by
up to 16 dB
Candidacy: Moderately-severe to profound, overlaps with CI
candidacy
Input-Output Functions
8000
DFC
4000
2.0
2.5 kHz
3.0
4.0
5.0
Freq.
2000
Level
All or none: only when needed the most
Spectral balance detector
Output Frequency (Hz)
Dynamic Linear Freq. Compression
Dynamic switching behavior
Level
Dynamic Linear Freq. Compression
2.5 kHz
Freq.
1000
500
250
250
500
1000
2000
4000
8000
Input Frequency (Hz)
7
Original
Dynamic Linear Freq. Compression
10k
9k
8k
7k
6k
Input
5k
ch il d r en l i ke s t rawberrie /z/
4k
3k
2k
1k
ch il d r en l i ke s trawberrie /z/
Output
6k
5k
4k
3k
2k
1k
Low-Pass Filtered @ 2.67 kHz
DFC, 3:1 ratio
ch il d r en l i ke s t rawberrie /z/
ch il d r en l i ke s t rawberrie /z/
Holes in the Audiogram
With precipitous losses,
detection of tonal sounds
might be very poor or
absent for frequencies <
2500 Hz where the hearing
drops out because
frequency-lowering switch
is not triggered
Detection of tonal sounds
> 2500 Hz significantly
improves again because
switch is activated
Linear Frequency Transposition
Filter
Start Freq.
“holes”
3536 Hz
source
Start frequency is start of inaudibility (e.g., dead region)
Source region starts about ½ octave below start freq.
Spectral peak detector: strongest peak in source region
= Aided Sound field
1768 Hz
target
Octave-wide band (re: target destination) is filtered around
source peak and copied (original peak is maintained) down one
octave
Adapted from: Widex (95020880001 #01)
8
Linear Frequency Transposition
‘Audibility Extender’ (AE)
2500
2.5 kHz
‘start’
10k
6 kHz
‘start’
Source
9k
*
7k
6k
5k
*
*
*
Input
*
Output Frequency (Hz)
8k
Actual start
4k
2000
3k
2k
1k
ch il d r en li ke strawberrie/z/ ee SH
Un-transposed
Basic AE
Expanded AE
1500
1000
750
6k
5k
?
4k
Output
3k
Target
500
500
1k
4 kHz Limit
3
2
Freq. Output, kHz
Freq. Output, kHz
4
1
2
3 4
5
Freq. Input, kHz
Start Freq.
Source
Region
Target
Region
Start Freq.
630
445-1260
223-630
800
566-1600
283-800
1000
707-2000
1250
3
2.5 kHz Limit
2
1
2
3 4
5
Freq. Input, kHz
6
Masking between the dual sources of information
How does the perceptual system differentiate between a transposed
high-frequency peak and an un-transposed low-frequency peak?
Dead regions
Profound loss in the high
frequencies with moderate loss in
the mid and low frequencies
Precipitous loss in high frequencies
Loss should not be too severe in
low frequencies, otherwise
cochlear distortion, along with high
gain and low MPO could further
distort the transposed signal
6000
Source
Region
Target
Region
630
891-1782
297-594
800
1131-2263
377-754
354-1000
1000
1414-2828
471-943
884-2500
442-1250
1250
1768-3536
589-1179
1600
1131-3200
566-1600
1600
2263-4525
754-1508
2000
1414-4000
707-2000
2000
2828-5657
943-1886
2500
1768-5000
884-2500
2500
3536-7000
1179-2333
3200
2263-6400
1131-3200
4000
2828-7000
1414-3500
6000
4242-7000
2121-3500
*Approximate Values: Transposed
signals extend beyond these regions
with decreasing amounts of gain
Linear Frequency Transposition
-10
Behavior is dynamic: exact frequency location of
transposed signal (source & target) varies
depending on the short-term input spectrum
Always active: may minimize signal discontinuities
and artifact
Harmonic relationship b/w original & transposed
signals maintained
0
10
20
30
Hearing Level, dB HL
4000
Expanded AE
Basic AE
6
Kuk et al. (2009)- only those with:
3000
4
Candidacy
2000
AE Frequency Reallocation Table
Concerns:
1500
5
1
1
1000
Basic AE: dominant peak shifted down by 1/2 (octave), source region begins ½
octave below start frequency
Expanded AE: dominant peak shifted down by 1/3, source region begins ½ octave
above frequency start; peak shift from basic AE remains
Linear Frequency Transposition
5
750
Input Frequency (Hz)
Source content
mixes with content
in target region
2k
40
50
60
70
80
90
100
110
120
250
500
1000
2000
Frequency, Hertz
4000
8000
May be more natural and pleasant
Pitch is unaltered
9
Linear Frequency Transposition
Nonlinear Frequency Compression
Only a portion of the spectrum around the peak is
transposed
4.0 kHz Start, 2.3:1 CR
Freq. Output, kHz
Limits the amount of masking and eliminates need for
frequency compression
High-frequency energy is often diffuse with a broad peak
potentially useful information about gross spectral
shape is discarded
1.5 kHz Start, 1.9:1 CR
5
Freq. Output, kHz
4
Start
3
2
1
5
4
3
2
Start
1
Transposed and un-transposed signals are mixed
Already mentioned issue with masking and segregation
Might transpose unwanted high-frequency background
noise and decrease SNR
1
Input
10
0
The higher the number, the more the input compression
band (CB) is condensed; very closely corresponds to the
bandwidth reduction on an ERBN scale:
Maximum
Input
Start
2000
4000
6000 8000
6
Actual numbers below:
20
15
ERBN Filter Bank
10
5
0
Frequency, Hz
0
Maximum
Output
2000
4000
Relative Magnitude, dB
≈ 4500 Hz
15
5
25
20
Relative Level, dB
Relative Level, dB
25
2
3
4
5
Freq. Input, kHz
CR ≠ in/out (on a linear Hz scale)
Output
1
Nonlinear Compression Ratio (CR)
30
Compression Band
6
Adjustable start frequency and compression ratio
How Nonlinear Freq. Comp. Works
30
2
3
4
5
Freq. Input, kHz
6000 8000
Frequency, Hz
-20
Input
Output
-40
-60
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
Frequency, Hz
Processing Limits: What You See
Lowest nominal start freq. = 1500 Hz
Processing Limits: What You Don’t See
Phonak CORE (pre-2011)
Highest nominal start freq. = 6000 Hz*
Input compression band is always < 4500 Hz
Max. input frequency is always ≤ start + 4500 Hz, but no
greater than ≈10,000 Hz* (6000 Hz in power device, Naída UP)
* 4000 Hz in power device (Naída UP)
Compression ratio = 1.5 – 4.0
Nonlinear on Hz scale (close to linear on ERBN scale)
Phonak Spice (2011)
Max. input frequency ≈10,000 Hz* (maybe limited by receiver output)
No additional limitation in frequency compression
bandwidth
10
NFC for Moderately Severe Loss
NFC for Mild-Moderate Loss
CORE Limits
CORE Limits
6k
9k
Source
8k
5k
Source
4k
7k
6k
Target
5k
3k
4k
*
2k
*
*
* *
*
Target
1k
ch
i l d
r
en
l i ke s t r aw b err ie
Flattened formant transitions
/z/
Spectral Feature Translation
ch
i l d
r
en
l i ke s t r aw b err ie
/z/
Spectral Feature Translation
Spectral Feature: algorithm uses a classifier that
looks for spectral features in the high-frequency
spectrum that are characteristic of speech
3k
2k
1k
BEFORE
Linear predictive coding: uses a source-filter model of
the speech production system to characterize the signal
Translation: features are synthesized at a lower
frequency
Modifies the filter by adding the high-frequency peak in a
lower frequency region
Uses the same source (i.e., harmonics are unaltered)
(Galster, personal communication)
Spectral Feature Translation
Spectral Feature Translation
10k
9k
8k
After
7k
6k
5k
4k
3k
2k
1k
ch il d r en l i ke s trawberrie /z/
6k
5k
4k
3k
2k
1k
11
Spectral Feature Translation
Like AVR and unlike Widex and Phonak,
frequency lowering is dependent on the input
Because lowering is triggered by a high-frequency
speech feature (i.e., frication), concerns about
transposition that is always active are minimized:
Masking and perceptual segregation of mixed signals
Lowering of unwanted high-frequency background
noise which can decrease SNR
Spectral Feature Translation
Like Widex in that a portion of the high frequency
information centered around the peak in the
target band is lowered
Does it in a way that maintains the harmonics of the
source band
Mixes the lowered and un-lowered spectra
Spectral Feature Translation
Unlike Widex and Phonak, does not roll off the
high-frequencies beyond the 2 upper channels
(> 5.7 kHz), which minimizes the risk that the
clinician might unintentionally limit audible
(Galster, personal communication)
bandwidth
Candidacy
Two Controls
1) Bandwidth of the source band
7 settings based on the high-frequency thresholds
2 defaults: a) severe-to-profound loss, b) moderate
hearing loss
Presumably, this affects the target band location
limit that peak information is lowered is somewhere
between 1000-1200 Hz
The mapping of source to target band has not been revealed
Lower
2) Gain of the lowered signal (0-10 dB)
1) Slope ≥ 25 dB per octave
2) Thresholds from 250-1000 Hz: ≤ 55 dB HL
3) Thresholds from 4000-8000 Hz: ≥ 55 dB HL
4) At least 1 threshold from 1000-3000: ≥ 55 dB HL
i.e., the mixing ratio of lowered and (Galster,
un-lowered
signals
personal communication)
Summary of Techniques
Activation
Technique
Recommended that all of the following be met:
Input Dependent
Always Active
Compression
AVR
(linear @ 0 Hz)
Phonak
(nonlinear @ start)
Transposition
Starkey
(feature lowering)
Widex
(peak lowering)
Research Findings
12
Effectiveness- A Few Points to Consider
Re-coded
information must go somewhere
Regions that would be otherwise amplified normally
Concern is not so much fidelity of re-coded information as it is newly
introduced distortion and sound quality in the regions it is moved to
A successful fitting requires giving the patient more than is taken away
Effectiveness- A Few Points to Consider
Information
Transmitted
Functional gain (aided threshold) does not equal
benefit
Moving everything to the low frequencies might
guarantee that it will be detected, but does not
guarantee that it will be discriminated
Re-coded speech must be learnable
Re-coded speech must retain a certain degree of sound
quality and naturalness
DO NO HARM
Depends on interaction
b/w hearing loss &
re-coding technique
Worse!
Amount of Speech Re-coding
(Distortion of Audible Bandwidth)
Effectiveness- A Few Points to Consider
While re-coded information may not precise, the
goal is to provide enough of it for top-down
processes to reduce uncertainty and use other
acoustic cues and context to fill in missing
content
Process of perceiving individual speech sounds
may be a matter of knowing more about what
they are not than what they are (Jakobsen &
Halle, 1956)
Effectiveness- More Points to Consider
Perhaps the focus should be individuals in
addition to group means
Interactions with audiometric configuration, peripheral
integrity, cognitive status, etc. are expected
Benefit may not always manifest as improvement
in speech intelligibility, but perhaps, ease of
listening and effort (like noise reduction),
production accuracy, etc.
Period of re-learning or acclimatization?
Dynamic Linear Freq. Compression
Dynamic Linear Freq. Compression
‘Meta-analysis’ by Simpson (2009)
Lack of homogeneity within and between studies
‘Meta-analysis’ by Simpson (2009)
Compared to conventional devices: about half of the
children (13/28) and one third of the adults (6/18)
demonstrated significant increases in speech
understanding with the DLFC, ranging from 10-20%
Devices varied from analog body worn to digital BTEs
High attrition rates due to consideration of CI or
cosmetic reasons
High repair rates were noted
Listeners with more severe loss seemed to benefit most
Those who were fit earlier and those who were in oral
communication programs seemed to benefit more
Possible confound with extra low-frequency gain
13
Frequency Transposition – Adults
Kuk et al. (2009)
8 adults with severe-to-profound loss > 2kHz
Micro BTEs, 20-30 minutes daily training for 1 month,
tested over 2 month period
After 2 months, fricative perception improvements of 510% increased to about 20%
Some decrement initially (confusion) for other
consonants (esp. stops and nasals)
Frequency Transposition – Children
10 children (6-13 yrs.) with severe-to-profound loss >
3kHz
6 weeks of training with transposition and 3 weeks
without
5-10% improvement in consonant identification (esp.
voiceless) at 50 dB HL presentation level in quiet
Poorest performers with FT de-activated had greatest
improvement when it was activated
Production accuracy of /s/ and /z/ improved by 10%
After 2 months either no difference (stops, nasals) or some
improvement < 10 % (affricates, glides, & liquids)
No degradation/improvement of performance in noise
Auriemmo et al. (2009)
Kuk et al. (2011): no training effect for control
Frequency Transposition – Children
Smith et al. (2009)
Nonlinear Freq. Comp. – Adults
Experimental body worn device for 4-5 weeks
17 experienced hearing aid users with moderate to
profound loss
Start frequency ranged from 1.6-2.5 kHz
Monosyllabic words: about half (8) had significant
improvement compared to conventional aids, 8 showed
no difference, 1 was worse
For the former, place of articulation and frication
improved over 10%
6 children (9-14 yrs.) with severe-to-profound loss > 1kHz
Improvements in CVC identification at 12 and 24 weeks
Poorly controlled study
“Strong acclimatization effect”
Simpson et al. (2005)
Proper counseling and training are highly recommended
Nonlinear Freq. Comp. – Adults
Simpson et al. (2006)
7 listeners with precipitous loss, start frequency = 1.01.6 kHz
Alexander: alteration of pitch and sound quality are associated
with these start frequencies
No significant improvement, except one in noise, some
new confusions
Compression parameters not individually chosen, 3
non-experienced users only wore device about 2
hours/week
Nonlinear Freq. Comp. – Adults
Glista et al. (2009)
Prototype BTE for 4 weeks (NFC activated/deactivated)
8 of 13 users with severe-to-profound loss showed
significant improvement, especially for plural
recognition, with no significant effect on vowel
perception
Start frequency ranged from 1.5 to 4.0 kHz
14
Nonlinear Freq. Comp. – Adults
Bohnert et al. (2010)
Nonlinear Freq. Comp. – Children
Glista
Commercial device compared to subjects own devices
7 of 11 listeners with severe to profound loss showed
some improvement for sentences in noise, 4 performed
more poorly
Acclimatization for subjective ratings (i.e., perceived
benefit, own voice, etc. continued to improve after 4
months of use)
Nonlinear Freq. Comp. – Children
Summary of Research Findings
Wolfe
et al. (2011)
children (5-13 yrs.) with moderate to moderatelysevere loss
6 months with commercial product aimed for children
Significant improvement after 6 weeks on plurals and
consonant identification, which further improved after
6 months
Improvement in noise (SRT) only after 6 months
No control or withdrawal condition cannot rule out
maturation
et al. (2009)
children (6-17 yrs.) with severe-to-profound loss
All had significant improvement on at least one speech
test, especially for plurals, with no significant effect on
vowel perception
Start frequency ranged from 1.5 to 3.5 kHz
11
Dynamic Linear Frequency Compression
15
Frequency Transposition
Severe-to-profound loss starting < 2.5 kHz
Normal to severe low-frequency hearing
Documented benefit for 1/2 of the children and 1/3 of the adults
Profound loss in the high freq. and moderate loss or better in low freq.
Should not be used to extend high-freq. BW for mild to moderate loss
Improvement in fricative perception in adults and children
Strong acclimatization effect lasting at least 2 months
Nonlinear Frequency Compression
Improvement for fricatives/affricates in adults and children with mild-tomoderate loss and moderate-to-profound loss
Exception for precipitous loss > 1.5 kHz where low start frequency create
more harm than benefit
Mild acclimatization for speech perception
0.90
Alexander et al. (2008)
NH
HI
Overall Percent Correct
Efficacy of frequency lowering > 4 kHz
24 adults with normal hearing
24 adults with mild to moderate loss
Only 11 were regular hearing aid users
Are error patterns a function of the frequency
lowering, the hearing loss, or both?
Proportion Correct
0.85
0.80
NH with
FT do
worse
than HI
without!
0.75
0.70
0.65
0.60
0.55
No FT
(No NR)
# of HI Listeners with
improvement (n=24)
No FT
(NR)
FT
WB (FT)
10
5
22
No FC
FC (10k)
FC (6k)
17
10
WB (FC)
15
15
Conclusions: Alexander et al. (2008)
Conclusions: Alexander et al. (2008)
Frequency transposition
Only apply to mild-moderate high-frequency loss
Substantially increased confusions, esp. [/f/, /th-/] and
[/v/, /TH+/]
For NH, place errors for /s/ and /z/ were new
Errors
Some form of FC above 3-4 kHz for HI listeners
with mild-moderate loss might be beneficial
are due to FT, not hearing loss
High start frequencies guarantee unaltered vowel
perception
Assumes that perception of other consonants and
fricatives/ affricates in other vowel contexts will be not
be degraded
Nonlinear frequency compression
NFC (6k) shows little harm/benefit compared to No NFC
for HI
75% of HI listeners benefited from NFC (6k) and/or NFC
(10k)
Alexander (2010) suggests this is true
Follow Up Study
12 HI listeners returned
3 conditions
1.0
No FC
FC (10k) in device
0.8
Follow Up Study
4.6 kHz start, 2.8:1 CR
Wideband
Significant improvement
for FC (10k) in device
compared to No FC
p < 0.01 (Wilcoxon)
Worst performers with No
NFC showed most benefit
No difference between NFC (10k) and wideband
(10k) NFC fully restored high-freq. information
1.0
0.6
0.5
0.4
0.3
p = 0.001
0.2
0.4
0.6
0.8
1.0
normal-hearing listeners
Consonants (n = 12)
0.6
0.5
0.4
0.2
72 nonsense syllables (h-V-d) presented in speech-shaped noise
at 6 dB SNR (Hillenbrand et al., 1995)
12 vowels
6 talkers (2 adult males, 2 adult females, 1 boy, 1 girl)
0.6
0.8
1.0
Alexander (2010) Summary
120 nonsense syllables (vCv) presented in speech-shaped noise at
10 dB SNR
20 consonants
3 vowel contexts (/a/, /i/, /u/)
2 talkers (adult male and female)
(n = 10)
0.4
WB (% Correct)
For all talkers and all speech classes, no start
freq. ≥ 2756 Hz showed a deficit
0.7
0.2
No NFC (% Correct)
12
Vowels
0.8
0.3
0.2
Alexander (2010)
NFC vs. WB
0.9
0.7
NFC (% Correct)
NFC (% Correct)
0.9
Little concern for using NFC when start frequency is
beyond the F2 range of most talkers (2067 – 2756 Hz)
When using low start frequencies be extra
attentive to user’s progress and counsel user
May need a period of acclimatization, even then, users
might still do worse
16
Alexander (2012): Research Question
Because the NFC start frequency and CR both
influence how frequencies are remapped, there
are infinite ways the unaidable high-frequency
spectrum can be repackaged into the audible
range of the listener
Lower start & lower CR
vs.
Low Start, Low CR
High Start, High CR
High Start, Low CR
Input Frequency
Input Frequency
Input Frequency
Output Frequency
Alexander (2012)
Higher start & higher CR
Alexander (2012): Research Question
Project examines the perceptual tradeoffs that
occur when trying to optimize the choice of NFC
start frequency and CR
Moderately-severe to profound high-frequency loss
Start Frequency
Hypothesis 1
Hypothesis 2
a) They allow a greater amount of high-frequency information
to be lowered if CR is fixed
b) They allow for lower CRs (less reduction in spectral
resolution) if the input bandwidth (the “max input” frequency)
is fixed
Mild to moderate high-frequency loss
Hypothesis 1
Listeners
CR should be increased so that a greater amount of highfrequency information can be lowered into the range of
audibility
Hypothesis 3
Group 1: Simulation of moderately-severe to
profound high-frequency loss
Hypothesis 2
CR should be kept low to maintain spectral resolution of the
lowered signal
The optimal CR depends on start frequency
E.g., lower CRs might be best for low start frequencies, but less
important for high start frequencies where formant frequencies are
less critical
Lower start frequencies are beneficial because
Compression Ratio
Lower start frequencies are detrimental for phonemes that
rely heavily on formant frequency, especially vowels
14 (6 male, 8 female) listeners with sensorineural loss,
ages 47-83 years (median = 70 years)
Group 2: Mild to moderate high-frequency loss
13 (6 male, 7 female) listeners with sensorineural loss,
ages 27-82 years (median = 62 years)
17
Conditions
1 control condition with no NFC
6 experimental conditions with NFC
Conditions
Moderately-severe to profound
Start =
1550 Hz
All test conditions were low-pass filtered to the
selected max output frequency
CR = 1.30
Start =
1550 Hz
CR = 1.58
Start =
1550 Hz
CR = 2.49
3273 Hz (Group 1)
4996 Hz (Group 2)
Max Input = 7063 Hz
Max Input = 9130 Hz
CR = Max Input =
1.52
4996 Hz
Start =
2239 Hz
CR =
2.01
A within-subjects, Latin Squares design was used
1
Can We Predict the Best Settings?
3
Max Input = 9130 Hz
4
5
6
Frequency, Hz
Start =
1550 Hz
CR = 1.57
Start =
1550 Hz
CR = 2.11
7
8
9
Max Input =
7063 Hz
Max Input = 9130 Hz
Start = 2756 Hz
CR = 2.03
Start = 2756 Hz
CR = 3.03
Start = 3962 Hz
Max Input = 7063 Hz
CR =
3.60
2
Max Output = 4996 Hz
Max Input =
4996 Hz
Start =
2239 Hz
Start =
2239 Hz
Mild to moderate
Max Output = 3273 Hz
Start = 3962 Hz
1
2
3
CR =
2.37
Max Input =
7063 Hz
Max Input = 9130 Hz
Max Input =
7063 Hz
CR =
3.70
Max Input = 9130 Hz
4
5
6
Frequency, Hz
7
8
Conclusions: Alexander (2012)
Most manufacturers of frequency-lowering hearing
aids recommend that the clinician adjust the setting
based the user’s perception of /s/-/ʃ/ via live voice
Improvements in fricative/affricate identification
should be expected when using NFC for a variety of
hearing losses
Listeners participated in an additional 1-hour task
that involved /s/-/ʃ/ discrimination for each of the
NFC conditions
For some settings this improvement might come at
the expense of a decrease in vowel and non-fricative
consonant identification
In practice, we may need to compromise on ideal
performance for fricatives/affricates to avoid a decrease in
performance for the vowels and other consonants
Perhaps, ideal performance for these settings will be
reached after a period of learning/acclimatization
The conditions that yielded best performance on this
task did NOT predict the conditions that yielded best
performance in the main experiment
Conclusions: Alexander (2012)
Low start frequencies should be avoided
In cases where the bandwidth of audibility is restricted,
it is better to tradeoff an increase in CR for a higher
start frequency
9
Conclusions: Alexander (2012)
While CR reduces spectral resolution of the lowered
signal, the important thing is where the compression
occurs and not so much how much of it there is
When the start frequency is low, increasing CR decreases
identification for vowels and non-fricative consonants
A slightly lower CR can be maintained by bringing less highfrequency information down into the range of audibility
However, if the reduction in high-frequency information is too
great, fricative/affricate identification might not be optimized
In cases where the bandwidth of audibility is less
restricted, attempts should be made to keep the start
frequency above the range of most second formants
For higher start frequencies, if sufficient high-frequency
information is brought down, CR seems to be less important
18
Conclusions: Alexander (2012)
Attempts to identify the best NFC setting on an
individual basis using a /s/-/ʃ/ discrimination task
or rules that simply maximize input bandwidth
are limited
Recommendations for selecting a setting also
need to consider the start frequency, the CR, and
the interaction between the two
To Fit or Not to Fit
Conclusion for today
There is not a lot of good evidence, period.
There is not a lot of evidence that existing techniques provide
benefit beyond improving fricative identification
There is some evidence that, when properly set, frequency
lowering does not do harm
Ultimately, a decision has to be made whether the
potential pros outweigh the cons
If the decision is to fit, then there are a few things you should
know about:
Be Your Own Best Guide
Special Verifit Test Signals
General misunderstanding by clinicians about how the
devices work and how to fit them
No standardized methods of fitting
Results might be highly individualized
“… four special versions of
the Speech-std (1) test
stimulus are provided in
Speechmap to assist in the
adjustment of frequencylowering hearing aids. These
are called Speech3150,
Speech4000, Speech5000
and Speech6300. The
Speech3150 stimulus has
had the bands at 1000 Hz
and above attenuated by 30
dB except for the 1/3 octave
band at 3150 Hz which is
unattenuated. [and so on].”
(www.audioscan.com/resources/audionotes/audionote2rev3lowrez.pdf)
Special Verifit Test Signals
FT Device
1000 Hz Start
How the technology of choice works
How much of lowered information the patient is getting (verification)
Frye FONIX FP35 v6.12
Presents a single (or double) tone to the hearing
aid and graphs the entire frequency response
AE off
AE (Basic)
AE (Expanded)
No evidence of the transposed energy in the “expanded” mode
19
SoundRecover Fitting Assistant
Introduced May 2009, downloaded by audiologists in at least
20 different countries worldwide (not authorized by Phonak)
Goal: maximize use of the audible bandwidth
Uses a fuzzy logic model for selecting the optimal
SoundRecover setting. Based on 4 variables:
Audible output bandwidth
Audible input bandwidth
Start frequency
Compression ratio
For your free copy, email:
[email protected]
For more information:
Video Tutorial: www.hearingseminars.com/p12364411
Slides: www.TinyURL.com/PurdueEAR (links to the above)
Family of Frequency I-O Curves
8000
7000
SoundRecover “off”
6.0k,1.5:1
Settings for one audiogram
from an actual CORE device
6.0k,2.1:1
5.5k,2.6:1
Output Frequency
SoundRecover Fitting Assistant
6000
4.4k,2.4:1
5000
3.6k,2.3:1
3.0k,2.2:1
4000
X
3000
*
*
2.6k,2.1:1
2.2k,2.0:1
1.9k,1.9:1
1.7k,1.8:1
1.5k,1.9:1
X
*
Max Audible
Output
1.5k,2.4:1
2000
1.5k,3.1:1
1.5k,4.0:1
1.5k,3.9:1
1000
0
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
Input Frequency
Maximum Audible Output Freq.
Enter Preset Options
For your free copy, email: [email protected]
Convenient Fitting Report
20
Identify the Optimal Setting
Optimal SoundRecover Setting
2.0 kHz start, 2.2:1
SR off
SR on
Setting B is chosen b/c of higher start; B provides 700 Hz more ‘information’ than A
Too Weak SoundRecover Setting
Stronger SoundRecover Setting
1.7 kHz start, 2.1:1
2.3 kHz start, 2.3:1
SR off
Optimal SR
Too Weak SR
SR off
Optimal SR
Too Weak SR
Stronger SR
Output is essentially identical as the
optimal setting, yet has 700 Hz less
information from the input!
Too Strong SoundRecover
MPO and SoundRecover
1.5 kHz start, 3.2:1
1/3 Octave Bands
SR off
Optimal SR
Too Strong SR
5
Expected Freq.
Actual Freq.
0
Under-utilized
Audible BW
Relative Level (dB)
-5
-10
-15
-20
-25
-30
-35
-40
-45
-50
250
500
1000
2000
4000
8000
Frequency (Hz)
21
MPO Comparisons
Binaural Fittings
Options:
Maximize information for each ear independently
SR off
2.0 kHz, 2.2:1
1.5 kHz, 3.2:1
(Different start frequencies)
Potential for introducing conflicting information between ears
Perhaps best when large differences in audible bandwidth are
due to large differences in hearing loss
Select the same setting for both ears
Select a compromise between the two ears
Select the optimal setting for the better ear
Perhaps best when small differences in audible bandwidth are
due to the fitting (e.g., feedback, insertion depth, RECD)
Audiogram 1
Audiogram 2
Default is
too strong
Audiogram 3
Default is
too weak
Audibility Extender Fitting Assistant
Goal: choose the setting that
Gives the most bandwidth …
with the least amount of overlap (highest start), …
and with no gaps (‘holes’) between un-transposed
and transposed signals
NFC is
inaudible
Adjust as necessary on follow up visits based
on client perceptions
Not authorized by Widex
[email protected]
22
1600 Hz Start
Overlap
2500 Hz Start
Ideal Scenario: Transposition
starts right when un-transposed
signal becomes inaudible
2000 Hz Start
Re-Scale
3200 Hz Start
“Hole”
(No audibility)
A Novel Way to Verify Frequency Shifting
Using a 3.5mm male-male phono cord extension (6
ft. is preferable), the headphone jack from the Verifit
is connected to the microphone jack on a standard
computer
Thank You!
www.TinyURL.com/PurdueEAR
For a full demonstration and free downloadable
software and test signal:
www.TinyURL.com/PurdueEAR
23
References
Alexander, J. M. (2009). “Candidacy, selection, and verification of SoundRecover options,” 3rd
Phonak Virtual Audiology Conference.
Alexander, J. M. (2010). “Maximizing benefit from nonlinear frequency compression,” 4th Phonak
Virtual Audiology Conference.
Alexander, J. M. (2012). “Nonlinear frequency compression: Balancing start frequency and
compression ratio,” American Auditory Society.
Alexander, J. M., Lewis, D. E., Kopun, J. G., McCreery, R. W., & Stelmachowicz, P. G. (2008). “Effects
of frequency lowering in wearable devices on fricative and affricate perception,” 2008 International
Hearing Aid Conference, Abs: 30-31.
American National Standards Institute. (2007). ANSI S3.5-1997 (Reaffirmed 2007), “American
national standard methods for the calculation of the speech intelligibility index,” New York (Author).
Auriemmo, J., Kuk, F., Lau, C., et al. (2009). “Effect of linear frequency transposition on speech
recognition and production of school-age children,” J. Am. Acad. Audiol. 20, 289-305.
Baer, T., Moore, B.C.J., & Kluk, K. (2002). “Effects of low pass filtering on the intelligibility of speech
in noise for people with and without dead regions at high frequencies,” J. Acoust. Soc. Am. 112,
1133-1144.
Best, V., Carlile, S., Jin, C. et al. (2005). “The role of high frequencies in speech localization,” J.
Acoust. Soc. Am. 118, 353-363.
Bohnert, A., Nyffeler, M., & Keilmann (2010). “Advantages of a non-linear frequency compression
algorithm in noise,” Eur. Arch. Otorhinolaryngol. 267, 1045-1053.
Brungart, D. S., & Simpson, B. D. (2009). “Effects of bandwidth on auditory localization with a noise
masker,” 126, 3199-3208.
Fox, R.A., & Nissen, S.L. (2005). “Sex-related acoustic changes in voiceless English fricatives,” J.
Speech, Lang., Hear. Res. 48, 753-765.
Füllgrabe, C., Baer, T., Stone, M. A., & Moore, B. C. J. (2010). “Preliminary evaluation of a method for
fitting hearing aids with extended bandwidth,” Int. J. Aud. 49, 741-753.
Glista, D. , Scollie, S., Bagatto, M., et al. (2009). “Evaluation of nonlinear frequency compression:
Clinical outcomes,” Int. J. Aud. 48, 632-644.
References
References
Moeller, M.P., Hoover, B., Putman, C., et al. (2007). “Vocalizations of infants with hearing loss
compared to infants with normal hearing: Part I- Phonetic development,” Ear Hear. 28, 605-627.
Moeller, M.P., McCleary, E., Putnam, C., et al. (2009). “Longitudinal development of phonology and
morphology in children with late-identified mild-moderate sensorineural hearing loss,” Ear Hear. 31,
625-635.
Moore, B.C.J. (2004). “Dead regions in the cochlea: Conceptual foundations, diagnosis, and clinical
applications,” Ear Hear. 25, 98-116.
Moore, B.C.J., & Tan, C-T. (2003). “Perceived naturalness of spectrally distorted speech and music,” J.
Acoust. Soc. Am. 114, 408-419.
Moore, B. C. J. et al. (2008); “Spectro-Temporal Characteristics of Speech at High Frequencies, and the
Potential for Restoration of Audibility to People with Mild-to-Moderate Hearing Loss,” Ear. Hearing,
907-9220.
Moore, B. C. J., Füllgrabe, C., & Stone, M. A. (2010). “Effect of spatial separation, extended bandwidth,
and compression speed on intelligibility in a competing-speech task,” J. Acoust. Soc. Am. 128, 360-371.
Moore, B. C. J., Füllgrabe, C., & Stone, M. A. (2011). “Determination of Preferred Parameters for
Multichannel Compression Using Individually Fitted Simulated Hearing Aids and Paired Comparisons,”
Ear Hear. 32, 556-568.
Pittman, A.L. (2008). “Short-term word-learning rate in children with normal hearing and children with
hearing loss in limited and extended high-frequency bandwidths,” J. Speech, Lang., and Hear. Res. 51,
785-797.
Pittman, A.L., Stelmachowicz, P.G., Lewis, D.E., & Hoover, B.M (2003). “Spectral characteristics of
speech at the ear: Implications for amplification in children,” J. Speech, Lang., Hear. Res. 46, 649-657.
Ricketts, T.A., Dittberner, A.B., & Johnson, E.E. (2008). “High-frequency amplification and sound quality
in listeners with normal through moderate hearing loss,” J. Speech, Lang., Hear. Res. 51, 160-172.
Grant, L., Bow, C., Paatsch, L., & Blamey, P. (2002). “Comparison of production of /s/ and /z/
between children using cochlear implants and children using hearing aids,” Poster presented at:
Ninth Australian International Conference on Speech Science and Technology (Melbourne).
Hamacher, J., Chalpupper, J., Eggers, E. et al. (2005). “Signal processing in high-end hearing aids:
State of the art, challenges, and future trends,” EURASIP J. App. Sig. Proc. 18, 2915-2929.
Hornsby, B. (2007). “Hearing aid bandwidth for speech understanding,” paper presented at: Am.
Acad. Aud. Convention (Denver).
Hornsby, B.W.Y., & Ricketts, T.A. (2006). “The effects of hearing loss on the contribution of highand low-frequency speech information to speech understanding. II. Sloping hearing loss.,” J. Acoust.
Soc. Am. 119, 1752-1763.
Hornsby, B.W.Y., Johnson, E.E., & Picou, E. (2011). “Effects of Degree and Configuration of Hearing
Loss on the Contribution of High- and Low-Frequency Speech Information to Bilateral Speech
Understanding,” Ear Hear. 32, 543-555.
Jakobson, R. & Halle, M. (1956). Fundamentals of Language. The Hague: Mouton
Johansson, B. (1961). "A new coding amplifier system for the severely hard of hearing," Proc. 3rd Int.
Congress on Acoustics, Stuttgart 1959, 2, 655-657.
Karlsen, B.L., Flynn, M.C., & Eneroth, K. (2006). “The benefit of high-frequency bandwidth for
hearing-impaired people under speech in noise conditions,” paper presented at: International
Hearing Aid Research Conference (Lake Tahoe).
Kuk, F., Keenan, D., Auriemmo, J., & Korhonen, P. (2009). “Re-evaluating the efficacy of frequency
transposition,” ASHA Leader, 14(1), 14-17.
Kuk, F., & Keenan, D. (2011). "Frequency transposition: Training is only half the story," Hearing
Review 17(12), 38-46.
Kuk, F., Keenan, D., Korhonen, P., & Lau, C. (2009). “Efficacy of linear frequency transposition on
consonant identification in quiet and in noise,” J. Am. Acad. Audiol. 20, 465-479.
References
Rudmin, F. (1983). “The why and how of hearing /s/,” Volta Review 85, 263-269.
Simpson, A. (2009). “Frequency-lowering devices for managing high-frequency hearing loss: A
review,” Trends Amplif. 13, 87-106.
Simpson, A., McDermott, H. J., & Dowell, R. C. (2005). “Benefits of audibility for listeners with severe
high-frequency hearing loss,” Hear. Res. 210, 42-52.
Simpson, A., Hersbach, A.A., & McDermott, H.J. (2005). “Improvements in speech perception with an
experimental nonlinear frequency compression hearing device,” Int. J. Aud. 44, 281-92.
Simpson, A., Hersbach, A.A., & McDermott, H.J. (2006). “Frequency-compression outcomes in
listeners with steeply sloping audiograms,” Int. J. Aud. 45, 619-629.
Sjolander, M. L., & Holmberg, M. (2009). “Broader bandwidth improves sound quality for hearingimpaired listeners,” Hear. Rev. 6, 40-45.
Smith, J., Dann, M., & Brown, M. (2009). “An Evaluation of Frequency Transposition for Hearingimpaired School-age Children,” Deafness Educ. Int. 11(2), 62–82.
Stelmachowicz, P.G., Hoover, B.M., Lewis, D.E., Kortekaas, R., & Pittman, A.L. (2000). “The
relationship between stimulus context, speech audibility, and perception for normal and hearing
impaired children,” J Speech, Lang., Hear. Res. 43, 902-914.
Stelmachowicz, P., Pittman, A., Hoover, B., & Lewis, D (2001). “The effect of stimulus bandwidth on
the perception of /s/ in normal and hearing-impaired children and adults,” J Acoust Soc Am;
110:2183-2190.
Vickers, D. A., Moore, B. C. J., & Baer, T. (2001). “Effects of low-pass filtering on the intelligibility of
speech in quiet for people with and without dead regions at high frequencies,” J. Acoust. Soc. Am.
110, 1164-1175.
Wolfe, J., Andrew, J., Schafer, E. C., Boretzki, M. & Caraway, T. (2009). “Long-term effects of nonlinear frequency compression for children with moderate hearing loss,” Int. J. Aud. 50, 396-404.
24