PRESTO: Preliminary Findings with a New High-Variability
Sentence Recognition Test in Patients with Cochlear Implants
1
1
1,2
2
Indiana University, Department of Psychological and Brain Sciences, Bloomington, IN., Indiana University School of Medicine, DeVault Research Laboratory, Indianapolis, IN.
120
100
60
40
20
0
Quiet
PRESTO in Quiet (RAU)
HINT in Quiet (RAU)
50
0
0
50
100
50
100
Lower scores indicate higher perceived handicap,
indicating that subjects reported the greatest
difficulty with scenarios related to speech perception,
and the least difficulty with quality of sound.
PRESTO
100
0
0
50
100
AzBio +10dBSNR (RAU)
50
Subject Gender Age
S1
S2
S3
S4
M
F
M
M
30
61
48
58
S5
S6
S9
S11
S12
S13
S14
S15
S16
S17
S18
S19
S20
M
F
F
M
F
M
F
M
M
F
M
M
M
73
71
55
77
44
54
29
83
92
50
74
77
86
Etiology
Otosclerosis
Unknown
Autoimmune
Genetic
Traumatic Head
Injury
Meniere's Disease
Unknown
Unknown
Meningitis
Spinal Meningitis
Unknown
Unknown
Genetic
Unknown
Viral - Sudden
Unknown
Meniere's Disease
Dur.
Severe
Loss
(years)
Dur. CI
Use
(years)
Device
2
12
1
10
2
1.5
19
12
Cochlear
Cochlear
Clarion
Cochlear
Bilat
CI L, HA R
CI L
CI R
0.5
5
3
5
2
3
9
12
4
6
0.5
4
20
5
2
10
2
9
22
7
2;0.5
5
2.5
0.5
5
0.75
Cochlear
Cochlear
MED-EL
Cochlear
ABC
Cochlear
Cochlear
Cochlear
Cochlear
ABC
ABC
Cochlear
MED-EL
Bilat
CI R
CI L
CI L
CI R
CI R
CI L
Bilat
CI L, HA R
CI R, HA L
Bilat
CI L
CI R, HA L
Config.
• PRESTO (Gilbert et al., 2013): Lists of 18
sentences from the TIMIT Sentence
database (Garafolo et al., 1993), each
list contains variation in syntactic structure
& length, key word familiarity, log frequency,
talker dialect & gender.
• HINT (Nilsson, Soli, & Sullivan, 1994): Lists of 10 sentences of 6-7 syllables with a simple, uniform
syntactic structure produced by a single male talker of General American English variety.
• Az-Bio (Spahr & Dorman, 2004): Lists of 20 sentences of conversationally produced speech from
four talkers (two male and two female) of General American English variety.
0
50
100
Auditory Working Memory Capacity (Crannell & Parrish, 1957)
SPAN for digits, letters, spondee, and monosyllabic words. Lists recorded by a single female talker,
presented in lists increasing in length.
The Speech, Spatial, and Qualities of Hearing Scale (SSQ - Gatehouse & Noble, 2004)
Self-report scale designed to assess auditory disability in a wide variety of listening situations.
49 questions cover many aspects of speech perception, spatial hearing, and more general qualities of
hearing. See example of one item on right.
Procedure
• Subjects were tested in a sound-treated booth, seated 1m from a single loudspeaker at 0° azimuth.
• All auditory stimuli presented at 65 dBA; noise condition: +10 dB SNR, 6-talker babble.
• Subjects were tested using their own clinical map on their preferred “everyday” setting with volume
set at a comfortable level.
• Presentation of sentence tests were randomized for each subject, one list per condition, quiet
preceded noise testing (6 lists total).
• For statistical analysis, percent correct scores were transformed to rationalized arcsine units (RAU)
(Studebaker 1985) to normalize variance across the range of scores and reduce ceiling effects.
• All noise conditions included two unscored practice sentences at +10 dB SNR.
• Auditory working memory span test presented at 65 dBA, 2 lists at each list length (2-10 items).
• Single-tem recognition testing followed SPAN.
• Subjects completed SSQ using paper/pencil format during testing session.
Author contact info: [email protected]
15
10
5
0
Digits
Letters
Words
Working Memory Capacity and Sentence Recognition
100
80
60
40
20
r= 0.639
p = 0.008
0
-20
0
20
40
60
Digit Span - Points
80
120
100
80
60
40
20
r= 0.619
p = 0.006
0
-20
100
0
20
50
100
SSQ - Speech
40
20
0
r=0.120
-20
150
0
50
100
SSQ - Spatial
150
20
0
Speech
Spatial
Qualities
100
80
60
40
20
0
r=0.387
-20
0
50
100
SSQ - Qualities
150
SSQ
Speech
Item 12
SSQ
Speech
Item 13
SSQ
Qualities
Item 10
40
60
Digit Span - Points
80
Question
(10 pt scale: “not at all” = 0 - “perfectly” = 10)
You are in a group of about five people, sitting
round a table. It is an otherwise quiet place. You
can see everyone else in the group. Can you
follow the conversation?
You are in conversation with one person in a room
where there are many other people talking.
Can you follow what the person
you are talking to is saying?
You are with a group and the conversation switches
from one person to another. Can you easily follow
the conversation without missing the start of what
each new speaker is saying?
Correlations
(n=16,*p<0.5; **p<0.005)
AzBio-N (r=.530*)
HINT-Q (r=.504*)
AzBio-N (r=.518*)
AzBio-N (r=.487*)
Can you easily have a conversation on the
telephone?
HINT-Q (r=.630**) & HINT-N (r=.715**)
AzBio-Q (r=.716**) & AzBio-N (r=7.08**)
PRESTO-Q (r=.701**) & PRESTO-N (r=.644**)
Do other people's voices sound clear and natural?
HINT-N (r=.608*)
AzBio-Q (r=.575*) & AzBio-N (r=.700**)
PRESTO-Q (r=.616*) & PRESTO-N (r=.715**)
CONCLUSIONS
1) As expected, sentence recognition performance was highly variable across
subjects. All subjects showed best performance on tests in quiet. Performance
decreased as a function of test material - HINT > AzBio > PRESTO.
2) PRESTO is feasible in high-performing CI patients and is a meaningful
assessment to add to the clinical toolbox. Additionally, PRESTO may be useful
for candidacy evaluations and to evaluate performance over time.
3) All measures of working memory capacity were strongly related to all
measures of sentence recognition, both in quiet and noise.
4) While the SSQ self-report total score did not correlate, individual
questions were reliable predictors of sentence recognition performance.
Spondees
120
r = 0.539
-20
SSQ
Speech
Item 11
As expected, scores decreased as a function of test
material, with digits > letters > words > spondees. All subjects,
even those with very poor word recognition, could complete
digit span testing.
20
0
SSQ
Speech
Item 3
Mean SPAN points for digits (n=17), letters (n=16), words (n=16),
and spondees (n=16). Error bars indicate 95% confidence intervals.
25
20
SSQ
ITEM
Figure 3. Mean SPAN Scores.
30
40
60
40
Table 2. Individual Item Analysis - SSQ and Sentence Recognition.
50
35
60
80
60
No significant correlations between self-report performance and any sentence
recognition test or noise condition. Individual items correlated strongly with several
sentence recognition tests and noise conditions, items and correlations listed in Table 2.
PRESTO +10dBSNR (RAU)
II. Working Memory Capacity
40
80
100
80
Performance on PRESTO in noise plotted as a function of self-reported responses to each subscale,
Speech (left), Spatial (middle), and Qualities (right). Solid lines indicate best-fit linear regression.
0
Sentence recognition was highly variable across test measures and noise conditions. Subjects performed best in
the quiet conditions for each of the three tests.
45
100
Figure 6. Self-Report and Sentence Recognition.
Performance shown for individual subjects on HINT, AzBio, and PRESTO (left to right), in quiet as a function of performance in
the noise condition (+10dB SNR ). Solid diagonal line represents equal performance across the two conditions. The filled symbol in
each plot represents the mean.
HINT in Noise +10dB SNR (RAU)
Table 1. Subject Demographics.
100
0
Figure 2. Individual Subject Performance on Sentence Recognition.
HINT in Quiet (RAU)
Sentence Recognition
100
HINT +10dBSNR (RAU)
Seventeen, post-lingually deafened, adult
cochlear implant users participated in this study.
Patient demographics provided in Table 1.
Materials
AzBio
HINT
Mean self report scores for each of three subscales: speech,
spatial, and qualities of hearing. Error bars indicate 95%
confidence intervals.
Mean performance decreased as a
function of test material, HINT > AzBio >
PRESTO in both quiet and noise conditions.
Noise
120
Total Score
80
Speech, Spatial, and Qualities Questionnaire
Figure 5. Mean Total SSQ Self-Report Scores.
A repeated measures ANOVA was used to test
for differences among test material and noise
condition. There was a significant main effect of
test condition, F(1,16) = 81.977 p < .001. There was
a significant main effect of noise condition, F(2,32)
= 46.307, p < .001. There was a significant
interaction between test material and noise
condition F(2,32) = 96.732, p < .001.
SPAN Points
Subjects
Mean performance on sentence tests for quiet
(left) and noise (right). Test material indicated by
color (see legend: HINT, AzBio, and PRESTO).
Error bars indicate 95% confidence intervals.
HINT
AzBio
PRESTO
The goals of the current study were:
1) To compare PRESTO with two conventional clinical tests of sentence
recognition (AzBio and HINT) presented in quiet and multi-talker babble.
2) To explore the relationship between demographic, cognitive, and
self-assessment measures with sentence recognition in order to characterize the
underlying information processing skills associated with speech perception in
adverse conditions.
METHODS
Figure 1. Mean Sentence Recognition.
III. Speech, Spatial and Qualities of
Hearing Scale (Gatehouse & Noble, 2004)
PRESTO in Noise (RAU)
I. Sentence Recognition
AzBio in Quiet (RAU)
• Speech in noise continues to be one of the most common problems facing
patients with cochlear implants.
• To perform well in everyday noisy environments, listeners must quickly adapt,
switch attention, and adjust to multiple sources of variability in both the signal
and listening environments.
• Sentence recognition tests in noise are useful for assessing speech recognition
abilities because they require a combination of basic sensory/perceptual abilities
as well as cognitive resources and processing operations.
• PRESTO (Perceptually Robust English Sentence Test Open-set) was
developed at IU as a sentence level test of speech perception that:
- More closely approximates real-world communication with multiple talkers.
- Engages more cognitive processing resources.
- Does not reach ceiling levels of performance in quiet, to allow for analysis of
individual differences.
- Differs from other sentence recognition tests because target sentences differ
in talker, gender, and regional dialect.
PRESTO in Noise (RAU)
RESULTS
PRESTO in Noise (RAU)
INTRODUCTION
RAU
R14
Kathleen F. Faulkner , Taylor Twiggs , David B. Pisoni
1,2
DEVAULT OTOLOGIC
RESEARCH LABORATORY
Selected References
100
Figure 4. SPAN and Sentence Recognition.
Performance on HINT in Quiet as a function of Digit Span (left panel) and HINT in Noise as a
function of Digit Span (right panel). HINT plotted as a representative example. Solid lines indicate best
fit regression, r and p values indicated in plots.
All working memory SPAN tests (digits, letters, words, and spondees) were highly
correlated with all sentence recognition tests both in quiet and noise.
•Crannell CW, Parrish, JM. (1957) A comparison of immediate memory span for digits, letters, and words. J Psychol: Interdisc and Applied 44:319-327.
•Garofolo JS, Lamel LF, Fisher WM, Fiscus JG, Pallett, DS & Dahlgren NL. (1993). The DARPA TIMIT acoustic-phonetic continuous speech corpus. Linguistic Data
Consortium, Philadelphia.
•Gatehouse S, Noble, W. (2004) The Speech, Spatial, and Qualities of Hearing Scale (SSQ). Int J Audiol 43:85-99.
•Gilbert JL, Tamati TN, Pisoni DB. (2013) Development, reliability, and validity of PRESTO: a new high-variability sentence recognition test. J Am Acad Audiol 24(1)
26-36.
•
Nilsson MJ, Soli SD, Sullivan J. (1994) Development of the hearing in noise test for the measurement of speech reception thresholds in quiet and in noise. J Acoust Soc
Am 95:1085-1099.
•
Spahr AJ, Dorman MF. (2004) Performance of subjects fit with the Advanced Bionics CII and Nucleus 3G cochlear implant devices. Arch Otolaryngol Head Neck
Surg.130:624-628.
•Studebaker GA. (1985) A “rationalized” arcsine transform. J Sp Hear Res. 28:455–462.
Work supported by NIH-NIDCD grants T32-DC00012 and R01-DC000111, and post-doctoral travel award.