Dr. Ayasakanta Rout - James Madison University
Dr. Ayasakanta Rout - James Madison University
Dr. Ayasakanta Rout - James Madison University
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
10/19/2012<br />
Coefficient of Agreement in Pairwise Comparison-<br />
Application in Hearing Aid Research<br />
<strong>Ayasakanta</strong> <strong>Rout</strong>, Ph.D.<br />
<strong>James</strong> <strong>Madison</strong> <strong>University</strong><br />
Lincoln Gray<br />
Lori Campanola<br />
Relevant financial disclosure: None<br />
Relevant non-financial disclosure: None<br />
of 50<br />
2<br />
of 50<br />
A word about hearing aid<br />
development lifecycle<br />
A word about hearing aid<br />
development lifecycle<br />
new DNR<br />
What we<br />
already know<br />
•Basic compression<br />
•Gain<br />
•Physical components<br />
•etc.<br />
new DFS<br />
Which New ones work<br />
developments (α-testing)<br />
new mic array<br />
new style<br />
What we<br />
already know<br />
Which New ones work<br />
developments (α-testing)<br />
Develop a hearing<br />
aid Beta-testing with those<br />
features<br />
3 of 50<br />
4<br />
of 50<br />
A word about hearing aid<br />
development lifecycle<br />
A word about hearing aid<br />
development lifecycle<br />
What we<br />
already know<br />
Which New ones work<br />
developments (α-testing)<br />
Develop a hearing<br />
aid Field with tests those<br />
features<br />
Marketing<br />
• In the absence of compelling evidence, and<br />
due to time pressure, it is not uncommon to<br />
launch a ‘novel’ feature in a hearing aid<br />
supported by ‘best guess’.<br />
• Unlike pharmaceutical research, there are not<br />
any potentially hazardous side effects of<br />
hearing aid signal processing.<br />
Typically 3 - 4 years and millions of $$<br />
5 of 50<br />
6<br />
of 50<br />
1
frequency<br />
10/19/2012<br />
Multichannel noise reduction<br />
• The incoming signal is split into multiple<br />
frequency channels<br />
• Signal to noise ratio in each channel is<br />
estimated<br />
– Frequency and depth of modulation<br />
– Overall amplitude of the signal across all channels<br />
– Co-modulation<br />
– Harmonic structure<br />
• Based on the analysis, gain reduction is<br />
Advantages of multichannel noise<br />
reduction<br />
• Reduces gain only in channels where<br />
predominantly noise is present.<br />
– Listening Comfort<br />
• Speech Intelligibility<br />
– Not enough evidence<br />
applied within individual channels 7 of 50<br />
8<br />
of 50<br />
Several approaches are available<br />
1. Temporal modulation based gain reduction<br />
2. Modified Weiner filtering<br />
3. Other forms of spectral subtraction<br />
Temporal waveform<br />
of speech and noise<br />
Speech<br />
Random noise<br />
9 of 50<br />
10<br />
of 50<br />
Harmonic structure of speech<br />
How can we measure the level of<br />
noise (gain) reduction<br />
Creating appropriate test stimuli<br />
time<br />
11 of 50<br />
12 of 50<br />
2
Amplitude<br />
10/19/2012<br />
Spectrogram of ICRA noise<br />
International Collegium of Rehabilitative Audiology<br />
ICRA noise with a two-octave notch<br />
centered around 2000 Hz<br />
13 of 50<br />
14<br />
of 50<br />
ICRA noise with the embedded<br />
steady state noise (two octave)<br />
ICRA noise with the embedded<br />
steady state noise ( 1 / 3 octave)<br />
15 of 50<br />
16<br />
of 50<br />
ICRA noise with the embedded<br />
steady state noise (one octave)<br />
3D graphical representation<br />
Frequency <br />
17 of 50<br />
18<br />
of 50<br />
3
Gain Reduction (dB)<br />
Gain Reduction (dB)<br />
Gain Reduction (dB)<br />
10/19/2012<br />
Methods<br />
Methods<br />
• Stimuli<br />
• Digital BTEs with multichannel noise reduction<br />
from five major manufacturers<br />
• Stimulus duration: 2 minutes each<br />
• Overall RMS amplitude equalized<br />
• Total 20 stimuli<br />
• Programmed for flat 60 dB HL<br />
• Directionality, feedback suppression, and<br />
manual volume control (where available)<br />
disabled.<br />
19 of 50<br />
20<br />
of 50<br />
Methods<br />
Manufacturer A<br />
65 dB SPL<br />
KEMAR<br />
with<br />
custom<br />
earmold<br />
ER-11<br />
with<br />
½ inch mic<br />
5<br />
0<br />
-5<br />
1/3-Octave 1-Octave 2-Octave<br />
Computer<br />
-10<br />
-15<br />
-20<br />
-25<br />
250 500 1000 2000 3000 4000<br />
Location of Band Rejection Notch (Hz)<br />
21 of 50<br />
22<br />
of 50<br />
Manufacturer B<br />
Manufacturer C<br />
5<br />
1/3-Octave 1-Octave 2-Octave<br />
1/3-Octave 1-Octave 2-Octave<br />
0<br />
5<br />
-5<br />
-10<br />
-15<br />
0<br />
-5<br />
-10<br />
-15<br />
-20<br />
-20<br />
-25<br />
250 500 1000 2000 3000 4000<br />
Location of Band Rejection Notch (Hz)<br />
-25<br />
250 500 1000 2000 3000 4000<br />
Location of Band Rejection Notch (Hz)<br />
23 of 50<br />
24<br />
of 50<br />
4
Gain Reduction (dB)<br />
Gain Reduction (dB)<br />
10/19/2012<br />
Manufacturer D<br />
Manufacturer E<br />
1/3-Octave 1-Octave 2-Octave<br />
1/3-Octave 1-Octave 2-Octave<br />
5<br />
250 500 1000 2000 3000 4000<br />
10<br />
5<br />
0<br />
0<br />
-5<br />
-5<br />
-10<br />
-15<br />
-10<br />
-15<br />
-20<br />
-20<br />
-25<br />
Location of Band Rejection Notch (Hz)<br />
-25<br />
250 500 1000 2000 3000 4000<br />
Location of Band Rejection Notch (Hz)<br />
25 of 50<br />
26<br />
of 50<br />
Ranking based on gain reduction<br />
D A B C E<br />
Part II: Paired Comparison of Sound<br />
Quality with Noise Reduction<br />
Better<br />
Worse<br />
27 of 50<br />
28 of 50<br />
• Are there any differences in sound qualities of<br />
noisy speech processed by these hearing aids<br />
– Paired comparison of Noisiness<br />
5 sentences<br />
from HINT<br />
@<br />
+5 dB SNR<br />
Manufacturer A<br />
Manufacturer B<br />
Manufacturer C<br />
Manufacturer D<br />
Manufacturer E<br />
5<br />
2<br />
= 10 pairs<br />
Methods<br />
• Recording HINT sentences<br />
– Five sentences presented at 65 dB SPL in sound<br />
field at +5 dB SNR<br />
– Output of hearing aids recorded on computer<br />
• Paired comparison<br />
– Every possible pairing of the five hearing aids<br />
presented<br />
29 of 50<br />
30<br />
of 50<br />
5
10/19/2012<br />
Subjects<br />
• Young, normal hearing subjects (air<br />
conduction thresholds at or better than 20 dB<br />
HL for 250-8000 Hz) and normal<br />
tympanograms<br />
• Age 18-30 yrs<br />
• N = 20 (12 M, 8 F)<br />
Paired Comparison of Noisiness Results<br />
• Hearing aids ranked in terms of less noisiness<br />
1. Manufacturer D<br />
2. Manufacturer A<br />
3. Manufacturer E<br />
4. Manufacturer C<br />
5. Manufacturer B<br />
31 of 50<br />
32<br />
of 50<br />
D<br />
A<br />
E<br />
C<br />
B<br />
What about the internal consistency<br />
within and across listeners<br />
Kendall and Smith (1939)<br />
Kendall and Gibson (1990)<br />
33 of 50<br />
34 of 50<br />
Circular Triads<br />
1 = Column chosen as “less noisy”<br />
0 = Row chosen as “less noisy”<br />
Hearing Aid<br />
A<br />
><br />
Hearing Aid<br />
B<br />
A<br />
A B C D E<br />
Hearing Aid<br />
B<br />
Hearing Aid<br />
C<br />
><br />
><br />
Hearing Aid<br />
C<br />
Hearing Aid<br />
A<br />
Inconsistent<br />
Response<br />
B 0<br />
C 0 1<br />
D 0 1 1<br />
E 0 1 1 1 _______<br />
35 of 50<br />
36<br />
of 50<br />
6
10/19/2012<br />
Coefficient of Agreement for the Reduction<br />
of Background Noise<br />
COA2 Analysis for Reduction of<br />
Background Noise Condition<br />
Number of<br />
Z Value<br />
Circular Triples<br />
Sentence 1 10<br />
13.8<br />
Sentence 2 5<br />
13.5<br />
Sentence 3 0<br />
14<br />
Sentence 4 6<br />
13.6<br />
Sentence 5 9<br />
13<br />
Subject Test-Retest Reliability for the<br />
Reduction of Background Noise Condition<br />
N = 6<br />
Test-Retest Reliability of Six Subjects for the Less Background Noise Condition<br />
Pearson Correlation Significance in Two-Tailed T-Test<br />
Sentence 1<br />
Sentence 2<br />
Sentence 3<br />
Sentence 4<br />
0.975<br />
0.989<br />
0.931<br />
0.98<br />
0.005<br />
0.001<br />
0.022<br />
0.003<br />
Sentence 5<br />
0.971<br />
0.006<br />
37 of 50<br />
38<br />
of 50<br />
What About<br />
Speech Intelligibility<br />
• Are there any differences in sound<br />
qualities of noisy speech processed by<br />
these hearing aids<br />
– Paired comparison of Speech Intelligibility<br />
39 of 50<br />
40<br />
of 50<br />
Methods<br />
• Same as “noisiness condition”<br />
• N=11 (6 M, 5 F)<br />
Paired Comparison of Speech Intelligibility<br />
• Hearing aids were ranked in terms of being<br />
“more intelligible” as<br />
1. Manufacturer C<br />
2. Manufacturer A<br />
3. Manufacturer E<br />
4. Manufacturer B<br />
5. Manufacturer D<br />
41 of 50<br />
42<br />
of 50<br />
7
10/19/2012<br />
C<br />
A<br />
E<br />
B<br />
D<br />
Coefficient of Agreement for the Speech<br />
Intelligibility<br />
COA2 Analysis for the<br />
Speech Intelligibility Condition<br />
Number of<br />
Z Value<br />
Sentence 1<br />
Circular Triples<br />
1<br />
6<br />
Sentence 2 2<br />
5.8<br />
Sentence 3 2<br />
6.5<br />
Sentence 4 3<br />
5.4<br />
Sentence 5 5<br />
5.7<br />
43 of 50<br />
44<br />
of 50<br />
Subject Test-Retest Reliability for the<br />
Speech Intelligibility Condition<br />
Ranking Summary<br />
Test-Retest Reliability of Six Subjects for the More Intelligible Condition<br />
Pearson Correlation Significance in Two-Tailed T-Test<br />
Sentence 1 0.951<br />
0.013<br />
Sentence 2 0.996 0<br />
Sentence 3 0.989 0.001<br />
Sentence 4 0.952 0.012<br />
Sentence 5 0.928 0.023<br />
Average gain<br />
reduction<br />
Best<br />
Worst<br />
D A B C E<br />
Less noisy D A E C B<br />
Speech<br />
intelligibility<br />
C A E B D<br />
45 of 50<br />
46<br />
of 50<br />
Conclusion<br />
• Anticipated that hearing aids that best enhance<br />
SNR would be judged as having more speech<br />
intelligibility<br />
– not supported by our data<br />
• Few circular triads and good test-retest<br />
reliability<br />
• Hearing Aid D preformed best in objective<br />
measurements but worse for speech<br />
intelligibility<br />
Conclusion<br />
• Hearing Aid A preformed well across all<br />
categories<br />
• Hearing Aid C and E did not provide much gain<br />
reduction for objective measurements but<br />
were higher ranked regarding speech<br />
intelligibility<br />
• Hearing aid B provided gain reduction only for<br />
low frequencies and ranked low for noisiness<br />
and speech intelligibility<br />
47 of 50<br />
48<br />
of 50<br />
8
10/19/2012<br />
Is there a correlation between the hearing aids<br />
that were judged to have less background noise<br />
and the hearing aids that were judged to have<br />
more speech intelligibility<br />
That’s all folks<br />
routax@jmu.edu<br />
Correlation Between Hearing Aids Being Judged as Having Less Background Noise<br />
and Being Judged as Having More Intelligibility for Eleven Subjects<br />
Pearson Correlation Significance in Two-Tailed T-Test<br />
Sentence 1 -0.256 0.677<br />
Sentence 2 -0.272 0.658<br />
Sentence 3 -0.258 0.675<br />
Sentence 4 -0.061 0.923<br />
Sentence 5 -0.088 0.888<br />
49 of 50<br />
50<br />
of 50<br />
9