1. 1Newman, R., & Chatterjee, M. (2013). Toddlers' recognition of noise-vocoded speech. The Journal of the Acoustical
Society of America, 133(1), 483-494.
2http://www.mrc-cbu.cam.ac.uk/improving-health-and-wellbeing/cochlear-implant/
3Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274(5294),
1926-1928.
4Miller, G. A., & Nicely, P. E. (1955). An analysis of perceptual confusions among some English consonants. The Journal
of the Acoustical Society of America, 27(2), 338-352.
Dr. Tina Grieco-Calub- PI, Dr. Casey Lew-Williams, Hillary
Snyder- Lab Mentor, Dr. Luke Flores- Bioscientist Program
Director, nuViBE, Howard Hughes Medical Institute
Acknowledgements
Ddd
-Participants in the clear speech condition performed significantly better on both the forced choice task and the syllable confusion
task than participants in the 8-channel vocoded speech condition, as seen in Figures 2 and 3.
-Figure 4 shows the relationship between performance on the forced choice task and the syllable confusion task for adults in the 8-
channel condition. Although not statistically significant, there is a trend emerging: adults who performed better on the syllable
confusion task (and thus those who were able to differentiate between phonemes), achieved greater scores on the foced choice
task. This suggests that spectral degradation may interfere with the tracking of TPs. More data are need to confirm this.
-Data from the Consonant Matrix (Table 2) shows that certain types of syllables are missed more than others in the syllable
confusion task when the artificial language is vocoded into 8 spectral channels. In Figures 5 and 6, listeners correctly identified
voiceless syllables (k,p,t) nearly every time, yet only correctly identified voiced syllables (b,d,g) 67% of the time. Additionally, back
(g,k) and liquid (r,l) syllables were missed 12.9% more than front (b,p) syllables and 24.1% more than middle (d,t) syllables.
-
-Cochlear implants (CIs) are prosthetic devices that provide
auditory perception to hearing impaired listeners (see Fig
1). 1 It is estimated that over 300,000 people worldwide
have received cochlear implants due to hearing loss.
Fig 1. An external and
internal view of a
cochlear implant.2
-Due to technical limitations, CIs can represent acoustic
information in only 12 to 22 independent frequency
channels, whereas the normal hearing (NH) ear can
represent thousands of frequency channels. As a result, CI
processing significantly degrades the spectral resolution of
sound. This spectral degradation can interfere with speech
understanding and, in children, the ability to learn their
native language.
-Adults and infants as young as 8-months use measurable
statistical relationships between neighboring speech
phonemes such as voicing, place, or articulation to identify
the boundaries of words.3 This is known as auditory
statistical learning.
-The transitional probability (TP) between sounds increases
the more consistently the sounds appear together.
pretty#baby
For example, the TP from pre to ty is greater than the TP
from ty to ba.
-The tracking of TPs relies on the ability to differentiate the
subtleties between phonemes. The goal of this research
project is to determine if spectrally degraded speech
hinders auditory statistical learning in NH adults.
Hypothesis: Spectral degradation implemented in CI speech
processors interferes with cues important for statistical
learning in language acquisition.
-Participants: Participants included 23 NH adults aged 18-
35 with English as their primary language.
-Stimuli: Artifical speech stream consisting of 4 trisyllabic
nonsense words repeated in random order (eg.
pabikugolatudaropitibudo…) that is unprocessed (clear
speech) or subjected to 8-channel noiseband vocoding.
Noiseband vocoding represents the envelope of the speech
signal in a pre-defined number of channels.
-Procedure: Familiarization Phase: Adults exposed to
artificial speech stream for 6 minutes.
-Test phase 1: consisted of the completion of a two-alternative forced-choice
task. After listening to two words, adults had to select the word that sounded
most familiar from the sound clip. Choices included words (TP=1), part words
(TP=.5), and non-words (TP=0), as seen in Table 1.
-Test Phase 2: consisted of a Syllable Confusion task. After listening to a certain
syllable, participants had to correctly identify it from a choice of 24.
Table 2. Consonant Matrix for syllable confusion task in 8-
channel vocoded condition.4
Fig 4. A correlation of participant scores on both Fig 5. Average scores for voiced and Fig 6. Average scores for front,
tasks in the 8-channel vocoded speech condition. voiceless syllables. middle, back, and liquid syllables.
Background Information
Methods
Conclusions
Results
Words Part words Non-Words
pabiku pigola robaku
golatu tudaro dolatu
daropi
tibudo
Stimulus Syllable
ResponseSyllable
b d g k l p r t Na
b 26 0 2 2 0 1 6 0 0
d 2 42 2 0 0 0 1 3 0
g 5 0 29 1 0 0 4 0 0
k 3 1 13 38 1 1 2 0 0
l 0 0 0 0 42 0 10 0 0
p 16 0 6 9 4 52 8 0 0
r 0 0 1 0 6 0 23 0 0
t 2 11 0 4 1 0 0 51 0
Na 0 0 1 0 0 0 0 0 0
0
10
20
30
40
50
60
70
80
90
100
Clear speech 8-Channel vocoded speech
Average%CorrectonForced
ChoiceTask
Type of speech
Forced Choice Task in Clear and 8-
Channel Vocoded Speech
Conditions
0
20
40
60
80
100
120
clear speech 8-Channel vocoded speech
Average%CorrectonSyllable
ConfusionTask
Type of Speech
Syllable Confusion Task in Clear
and 8-Channel Vocoded Speech
Conditions
N=14 N=9
0
10
20
30
40
50
60
70
80
90
100
Voiced Voiceless
%Correct
Voicing
Voicing in Syllable Confusion Task
0
20
40
60
80
100
120
Front Middle Back Liquids
%Correct
Placement
Placement in Syllable Confusion
Task
y = 0.2546x + 36.824
0
10
20
30
40
50
60
70
80
0 10 20 30 40 50 60 70 80 90
%CorrectonForcedChoiceTask
% Correct on Syllable Confusion Task
Syllable Confusion vs Forced Choice in 8-Channel
Vocoded Speech Condition
Fig 2. Average scores for participants
in the the forced choice task.
Fig 3. Average scores for participants
in the syllable confusion task.
* *
Table 1. Words, part words, and non-words used in the study.
Citations
p < 0.001
N=14 N=9
p < 0.001
Editor's Notes
-start out with saying Cis provide access to individuals who are deaf, although these devices provide access to sound, the signals they provide are highly spectrally degraded
As a result, CI users don’t have access to the high fidelity frequency cues that individuals with NH have
How do children break into their native language
Adults have experience with language
-what results to include
-what pictures to include
-what to call test phase 2
-what tests were performed on data
-define what cochlear implants are
-define vocoded in the methods section
In the familiarization phase, adults were exposed to a 6-minute, pause free, artificial speech stream consisting of 4 tri-syllabic nonsense words repeated in random order.
Speech sample: pabikugolatudaropi…
CI users don’t have access to the high fidelity frequency cues that normal hearing (NH) individuals have because the signal that these devices provide are highly spectrally degraded. As a result, CI users are generally slower at processing and using speech compared to their NH counterparts.
The correlation in Figure 4 shows that listeners who performed better on the syllable confusion task (and thus those who were better able to differentiate between phonemes), achieved greater scores on the forced choice task. This suggests that spectral degradation does interfere with the tracking of TPs.
-Data from the Consonant Matrix (Table 2) shows that certain types of syllables are missed more than others in the syllable confusion task during the 8-channel vocoded condition. As seen in Figures 5 and 6, listeners correctly identified voiceless syllables (k,p,t) nearly every time, yet only correctly identified voiced syllables (b,d,g) 67% of the time. Additionally, back (g,k) and liquid (r,l) syllables were missed 12.9% more than front (b,p) syllables and 24.1% more than middle (d,t) syllables.