| dc.description.abstract |
<p>Cochlear implant listeners are presented with a time and frequency-quantized version
of speech signals. In the frequency domain, resolution is limited by the number of
electrodes in each listener's array. Current cochlear implant speech processing strategies
implicitly assume that the information presented to each one of these electrodes is
perceived as unique and independent. However, previous research suggests that stimuli
presented on different electrodes can be indiscriminable (e.g. Zwolan et al., 1997;
Throckmorton and Collins, 1999; Henry et al., 2000) . Additional studies suggest that
stimuli presented on one electrode can influence the perception of stimuli on neighboring
electrodes (e.g. Shannon, 1990; Chatterjee and Shannon, 1998; Boëx et al., 2003).
Removing this redundant or occluded information could cause more distinct or perceivable
information to be presented to the listener and possibly result in improved speech
recognition.</p><p>Previous studies have used psychophysical data to identify the
electrodes with the highest potential to confound speech recognition (Zwolan et al.,
1997, Boëx et al., 2003, and Garadat et al., 2012). In order to minimize electrode
interactions and maximize the amount of perceivable information, each of these studies
used a single psychophysical metric to deactivate the electrodes across all time windows
of the speech processing strategy. For some listeners, these reduced electrode sets
resulted in improved speech recognition over using the of the electrodes in their
array. These studies did not compare the results of using different psychophysical
metrics to exclude electrodes for a group of listeners nor did they investigate speech
recognition performance as a function of the number of electrodes excluded from the
array.</p><p>In this work, three different psychophysical metrics were used to obtain
a multidimensional estimate of the potential "usefulness'' of each electrode. These
results were then used to inform two different methods of psychophysics-motivated
electrode selection. The first method incorporated individual data into each listener's
energy-driven speech processing strategy. For each time window, the electrodes with
the highest energy that were also most likely to be perceived, according to the psychophysical
data, were selected for stimulation. The second method sequentially excluded the electrodes
with the highest potential to confound from the array across all time windows, resulting
in a group of psychophysics-motivated electrode sets for each metric. Evaluating each
of these electrode sets exhaustively would require a prohibitive amount of experimental
time. To mitigate this problem, an adaptive procedure was developed to estimate performance
as a function of cochlear implant parameters in a time-efficient manner. For each
metric, the procedure estimated the set with the highest estimated probability of
correct phoneme identification. Listeners' speech recognition performance using this
electrode set was then compared to their performance using their full electrode array.
For both electrode selection methods, listeners' speech recognition scores were generally
comparable to those obtained in the clinical condition. This finding supports the
hypothesis that listeners were not perceiving all the information presented to them
using their clinical speech processing strategy and their complete set of electrodes.
Additionally, these results suggest that improvements to the proposed electrode selection
strategies should be in investigated in order to increase the amount of perceivable
information presented to cochlear implant listeners.</p>
|
|
