April 1 through June 30, 2000
NIH Project N01-DC-8-210
Speech Processors for Auditory Prostheses
Prepared by
Blake Wilson, Robert Wolford and Dewey Lawson
Center for Auditory Prosthesis Research
Research Triangle Institute
Research Triangle Park, NC 27709
CONTENTS
III. Plans for the next quarter
V. Erratum for Quarterly
Progress Report 6
Appendix 1: Summary of reporting activity for this quarter
Appendix 2: Supplemental
data for Section II
The main objective of this project is to design, develop, and evaluate speech processors for implantable auditory prostheses. Ideally, such processors will represent the information content of speech in a way that can be perceived and utilized by implant patients. An additional objective is to record responses of the auditory nerve to a variety of electrical stimuli in studies with patients. Results from such recordings can provide important information on the physiological function of the nerve, on an electrode-by-electrode basis, and also can be used to evaluate the ability of speech processing strategies to produce desired spatial or temporal patterns of neural activity.
Work in this quarter included:
II. Further studies to evaluate effects of changes in stimulus rate and envelope cutoff frequency for CIS processors
In Quarterly Progress Report 6 (QPR 6) for this project we described studies to evaluate effects of changes in stimulus rate and envelope cutoff frequency for CIS processors. The studies included (1) tests of consonant identification with a recorded male talker, for many combinations of rate and envelope cutoff frequency in 4- or 6-channel CIS processors, for each of four subjects; (2) tests of sentence recognition for seven combinations of rate and envelope cutoff frequency in a single-channel variation of CIS processors, for one subject; and (3) tests of consonant identification with male and female talkers, monosyllabic word recognition, and sentence recognition for various rates of stimulation while holding the envelope cutoff frequency constant, for each of four subjects. In broad terms, the results showed that large increases in speech reception performance can be obtained with increases in rate up to around 500 pulses/s/channel for multichannel processors, and with increases in envelope cutoff frequency up to 50 to 200 Hz using those same processors. Some subjects enjoyed further significant gains in performance with further increases in rate; indeed, with rare exception the highest scores were achieved by each subject at the highest rate used for each test. Results from the tests with the one subject using a single-channel processor indicated a significant advantage of an especially high rate (10162 pulses/s) compared with lower rates (833 and 2525 pulse/s). These latter results were obtained in the absence of electrode interactions, which may increase with increases in rate for multichannel processors.
In the present report we describe further studies involving manipulations in rate and envelope cutoff frequency in multichannel CIS processors, over wide ranges of both parameters. We have extended the prior studies by including additional tests in quiet for two of the subjects and by including tests in noise for another subject. The new studies include (1) tests of consonant identification for a recorded female talker, for two of the subjects previously tested with the male talker; (2) tests of sentence recognition for those same two subjects; and (3) tests of consonant identification in noise using the male talker, for another of the previously-tested subjects.
Methods
Many of the methods used for the prior studies also were applied for the present studies. In particular, the processors, stimulation equipment, test materials, and procedures for testing were the same across the two sets of studies. These methods are described in QPR 6.
Unlike the studies described in QPR 6, the present studies included presentation of consonants in competition with noise. CCITT noise was used, which has a spectrum that matches the long-term spectrum of speech.
Subjects SR2, SR9 and SR10 participated in the present studies. Characteristics of these subjects, such as levels of performance with their everyday speech processors, are presented in QPR 6.
Results
As noted above, the present studies included evaluation of possible talker effects in tests of consonant identification, measures of sentence recognition for the same processor conditions and subjects, and evaluation of effects of noise interference on consonant identification for a broad range of processor conditions.
Multiple talkers. Results from tests of consonant identification
with male and female talkers are presented in Fig. 1, which shows interpolated
contour plots of the percent correct scores for each subject and talker
condition and for the two talkers combined. The percent correct scores
underlying these plots, along with the standard error of the mean for each
score, are presented in Appendix Fig. A.1.
Fig. 1. Consonant identification for the male,
female and combined talkers. Data for subject SR9 are from a 16 consonant
test, and data for subject SR10 are from a 24 consonant test. Percent correct
scores are shown as points; their projections to the rate/cutoff frequency
plane and interpolated and extrapolated contour plots are provided as well.
Matrices of the percent correct scores and associated standard errors of
the means are presented in Appendix Fig. A.1. Note that the ranges of rates
and envelope cutoff frequencies (LPF cutoff) are different between the
two subjects.
The plots in Fig. 1 include contours over all possible combinations of rate and envelope cutoff frequency, for the indicated ranges of those parameters. However, combinations with cutoff frequencies above the rate of stimulation were not tested for subject SR9, and combinations with cutoff frequencies higher than one-half the rate of stimulation were not tested for subject SR10. Thus, the rightmost corner in each of the contour plots is extrapolated far beyond the data. Individual data points and their projections to the rate/cutoff frequency plane are provided to aid the reader in considering the contour shapes.
The ranges of tested rates and cutoff frequencies differed between the two subjects. Measurements with SR9 included rates from 50 to 2525 pulses/s/channel and cutoff frequencies from 12.5 to 1600 Hz, whereas the ranges for SR10 included rates from 100 to 1634 pulses/s/channel and cutoff frequencies from 50 to 800 Hz.
As described in QPR 6, the contour plots for the male talker convey an impression of decrements in percent correct scores with reductions in rate of stimulation, or envelope cutoff frequency, or both, below certain values. Also, both subjects appear to have a peak in scores at particular combinations of rate and cutoff frequency. Subject SR10, for instance, obtained his highest score for the male talker with the highest tested rate (1634 pulses/s/channel) and the cutoff frequency of 200 Hz.
The contour plots for the female talker resemble those for the male talker, for both subjects. Combinations of low rates and low cutoff frequencies produce decrements in performance, as is the case for the male talker. The highest scores for both subjects are obtained with the highest tested rate of stimulation and at intermediate cutoff frequencies.
Although the plots for the male and female talkers are broadly similar for both subjects, they are different in some ways as well. For subject SR9, scores with the female talker are lower than the scores with the male talker. Also, manipulations in rate and cutoff frequency do not appear to produce effects with the female talker that are as large as those observed with the male talker. For SR10, scores with the female talker are relatively uniform over a large portion of the parametric space, compared with the pattern of scores for the male talker. Scores obtained with the female talker generally approximate or exceed scores obtained with the male talker over the tested ranges of rates and cutoff frequencies for this subject.
The patterns for the combined talkers indicate (1) relatively high scores for the highest tested rates and intermediate cutoff frequencies for both subjects, and (2) decrements in performance with combinations of low rates and low cutoff frequencies.
Quantitative comparisons among the scores are presented in Fig. 2, which
shows difference matrices for both subjects and for the male, female and
combined talkers. Each matrix was derived by first calculating the mean
of all percent correct scores for the selected subject and talker, and
then subtracting the mean from each percent score in the percent correct
matrix for that subject and talker (see Appendix Fig. A.1).
Fig. 2. Difference matrices for tests of consonant
identification with the male and female talkers and with both combined,
subjects SR9 and SR10. Each matrix was derived by subtracting the mean
of all percent-correct scores for that matrix (see Appendix Fig. A.1) from
the percent-correct score in each cell. A one-way analysis of the variance
(ANOVA) was conducted to evaluate the possibility of significant differences
among the scores for each matrix. The ANOVAs indicated significant differences
for all six matrices in this figure (and the corresponding matrices in
Fig. A.1). Shaded areas in each matrix above indicate scores that are significantly
different from the mean according to the Fisher least-significant-difference
(LSD) criterion (p < 0.05), and the bold type indicates scores
that are significantly different from the mean according to the more-conservative
Tukey honestly-significant-difference (HSD) criterion (p < 0.05).
Light shading indicates scores significantly below the mean, and dark shading
indicates scores significantly above the mean. Underlines indicate the
minimum and maximum values in each matrix. Scores within the heavy lines
(for subject SR9) were obtained with 20 blocks of the consonant tests for
the male and female talkers, and with 40 blocks for the combined talkers.
Scores outside the lines for SR9, and all scores for SR10, were obtained
with 10 blocks each for the male and female talkers, and with 20 blocks
for the combined talkers.
A one-way analysis of the variance (ANOVA) was conducted for each of the matrices in Fig. 2 and in each case indicated significant differences among the difference scores (or the raw scores, which would produce the same result in a one-way ANOVA). The shaded areas in each matrix identify processor conditions that produced scores that are significantly different from the grand mean of all scores according to the post-hoc Fisher least-significant-difference (Fisher LSD) criterion (p < 0.05), and the bold type identifies conditions that produced scores that are significantly different from the mean according to the Tukey honestly-significant-difference (Tukey HSD) criterion (p < 0.05). The light shading indicates scores significantly below the mean, and the dark shading indicates scores significantly above the mean. The underlined type indicates the maximum and minimum values in each matrix. Note that the Tukey HSD criterion is more conservative than the Fisher LSD criterion (i.e., a greater number of scores are identified as significantly different from the mean using the Fisher LSD criterion).
In all but one of the matrices, scores that are significantly below the mean are restricted to the lowest tested rate, the lowest tested cutoff frequency, or both. In the matrix for subject SR9 and the male talker, a score significantly below the mean also is found at the relatively low rate of 100 pulses/s/channel and the cutoff frequency of 100 Hz. Conversely, scores that are significantly above the mean are obtained with either relatively high rates (male talker, subject SR9) or the highest tested rate (all other matrices), combined with intermediate cutoff frequencies.
The Fisher LSD and Tukey HSD for post hoc comparisons in each
matrix of Fig. 2 are presented in Table 1. These values allow comparisons
of scores between any two cells within one of the matrices. Differences
in scores that exceed the Fisher LSD are significant according to that
criterion, and differences that exceed the Tukey HSD are significant according
to that more-conservative criterion.
Table 1. Critical values for comparisons between scores within
matrices in Figs. 2 and A.1.
| Fisher LSD | Tukey HSD | ||||||
| Subject | Talker(s) | outsidea | acrossb | withinc | outsidea | acrossb | withinc |
| SR9 | Male | 9.85 | 8.53 | 6.97 | 19.69 | 17.05 | 13.92 |
| Female | 10.74 | 9.30 | 7.59 | 21.46 | 18.59 | 15.18 | |
| Both | 9.66 | 8.37 | 6.38 | 19.26 | 16.68 | 13.62 | |
| SR10 | Male | 8.06 | 14.08 | ||||
| Female | 6.49 | 11.33 | |||||
| Both | 5.79 | 10.06 |
Another view of the results is presented in Figs. 3 and 4. These figures
show slices through the rate/cutoff frequency data for each of the subjects,
respectively. The left column in each figure shows slices along rows in
the matrices of the percent correct scores (Fig. A.1), and the right column
in each figure shows slices along columns in the matrices. Plots in the
left columns of the figures thus indicate the dependence of scores on the
cutoff frequency, with rate of stimulation as the parameter, and the right
columns indicate the dependence of scores on the rate of stimulation, with
cutoff frequency as the parameter. Standard errors of the means for the
scores presented in Figs. 3 and 4 can be found in Fig. A.1. The standard
errors ranged from 1.3 to 4.9 percent for subject SR9, and from 1.3 to
4.0 percent for subject SR10.
Fig. 3. Slices through the rate/cutoff frequency
data for subject SR9, from tests of consonant identification with male
and female talkers and with both combined. The left column shows slices
along rows in the matrices of the percent correct scores for this subject
(Fig. A.1), and the right column shows slices along columns in the matrices.
Plots in the left column indicate the dependence of scores on cutoff frequency
while holding the rate of stimulation constant, whereas plots in the right
column indicate the dependence of scores on rate of stimulation while holding
the cutoff frequency constant. Standard errors of the means for the scores
ranged from 1.3 to 4.9 percent and are presented in Fig. A.1. The legend
for the left column indicates rates of stimulation in pulses/s/channel
(pps/ch), and the legend in the right column indicates envelope cutoff
frequencies in Hz (LPF).
Fig. 4. Slices through the rate/cutoff frequency
data for subject SR10, from tests of consonant identification with male
and female talkers and with both combined. As in Fig. 3, the left column
shows slices along rows in the matrices of the percent correct scores for
this subject (Fig. A.1), and the right column shows slices along columns
in the matrices. Standard errors of the means for the scores ranged from
1.3 to 4.0 percent and are presented in Fig. A.1. The gray symbols and
lines show scores for additional conditions tested with the male talker
for this subject.
Additional data collected in tests with SR10 using the male talker are shown for reference by the gray symbols and lines in Fig. 4. These are the data that were reported in QPR 6 for this subject and talker. (The range of cutoff frequencies covered for the female talker is smaller than that covered for the male talker for this subject.)
The plots in Fig. 3 appear to show a general increase in scores with increases in cutoff frequency or rate. Scores for the cutoff frequency of 12.5 Hz, for instance, appear to be lower than the scores for higher cutoff frequencies, at least for the male talker. Scores appear to increase over the range of tested rates for the male talker, with the largest increase at the increment in rate from 50 to 100 pulses/s/channel. Scores for the female talker appear to increase with increases in rate beyond 417 or 833 pulses/s/channel. Scores for the combined talkers appear to increase over the tested range of rates.
The plots in Fig. 4 do not seem to show a sensitivity to changes in cutoff frequency for the fixed rates above 100 pulses/s/channel and over the range of cutoffs tested with both talkers (from 50 to 800 Hz). With the inclusion of the additional data for the male talker, scores appear to rise with increases in cutoff frequency from 12.5 to 50 Hz. In that case, increases in rate (right column) also appear to produce increases in scores, at least from 100 to 200 pulses/s/channel and the cutoff frequency conditions common to both talkers. Inclusion of the additional data for the male talker appears to diminish the effect of increases in scores with increases in rate.
Overall effects of changes in cutoff frequency and rate of stimulation
are illustrated in Fig. 5 for subject SR9 and in Fig. 6 for subject SR10.
The plots in the left columns of these figures show the average of scores
for all tested rates of stimulation at and above 200 pulses/s/channel (scores
from the left columns of Figs. 3 and 4 versus cutoff frequency), and the
plots in the right column show the average of scores for all tested cutoff
frequencies at and above 25 Hz (scores from the right columns of Figs.
3 and 4 versus rate of stimulation). Lower rates and lower cutoffs were
excluded in the derivation of plots for the left and right columns, respectively,
because these conditions produced clearly lower and sometimes different
patterns of scores compared to the included conditions. (This procedure
also was followed in the derivation of similar figures for QPR 6.) The
numbers in the bottom panels indicate how many scores were included in
the calculations of the average values. The error bars show standard errors
of the means. The gray symbols and lines in Fig. 6 show results from inclusion
of the additional cutoff frequency conditions used in the prior tests with
subject SR10.
Fig. 5. Averages of scores derived from data in
corresponding panels of Fig. 3, for subject SR9. The plots in the left
column of the present figure show the averages of scores for all tested
rates of stimulation at and above 200 pulses/s/channel, and the plots in
the right column show the averages of scores for all tested cutoff frequencies
at and above 25 Hz. The numbers in the bottom panels for each column indicate
the number of scores included in calculations of the means for each cutoff
frequency (left column) or each rate (right column). The bars show standard
errors of the means.
Fig. 6. Averages of scores derived from data in
corresponding panels of Fig. 4, for subject SR10. As in Fig. 5, plots in
the left column of the present figure show the averages of scores for all
tested rates of stimulation at and above 200 pulses/s/channel, and the
plots in the right column show the averages of scores for all tested cutoff
frequencies at and above 25 Hz. The numbers in the bottom panels for each
column indicate the number of scores included in calculations of the means
for each cutoff frequency (left column) or each rate (right column). The
bars show standard errors of the means. The gray symbols and lines show
results with the inclusion of additional data collected for this subject
in tests with the male talker (see Fig. 4). The gray numbers in the top
panels indicate the number of scores included in the calculations of the
means with the additional data.
A one-way ANOVA was conducted for the data in each of the panels in
Figs. 5 and 6. The results are presented in Table 2 for subject SR9 (corresponding
to Fig. 5) and in Table 3 for subject SR10 (corresponding to Fig. 6).
Table 2. Significant differences among scores for the cutoff
frequency and rate of stimulation conditions presented in Fig. 5 (data
for subject SR9). The differences indicated are those identified by the
Fisher LSD criterion (p < 0.05). Entries for cutoff frequencies
are in Hz, and entries for rate of stimulation are in pulses/s/channel.
| Talker(s) | Manipulation | ANOVA | Significant Differences |
| Male | Cutoff frequency | p < 0.005 | 1600, 800, 400, 200, 100, 50, 25 > 12.5 |
| Female | NS | ||
| Both | p < 0.05 | 1600, 800, 400, 200, 100, 50, 25 > 12.5 | |
| Male | Rate | p < 0.001 | 1634 > 417, 100, 50 |
| 2525, 833, 417, 200, 100 > 50 | |||
| Female | p < 0.001 | 2525 > 1634, 833, 417, 200, 100, 50 | |
| 1634 > 417 | |||
| Both | p < 0.001 | 2525 > 833, 417, 200, 100, 50 | |
| 1634 > 417, 100, 50 | |||
| 833 > 417, 50 | |||
| 417, 200, 100 > 50 |
Table 3. Significant differences among scores for the cutoff
frequency and rate of stimulation conditions presented in Fig. 6 (data
for subject SR10). The differences indicated are those identified by the
Fisher LSD criterion (p < 0.05). Entries for cutoff frequencies
are in Hz, and entries for rate of stimulation are in pulses/s/channel.
| Talker(s) | Manipulation | ANOVA | Significant Differences |
| Male | Cutoff frequency | NS | |
| Female | NS | ||
| Both | NS | ||
| Male | Rate | p < 0.05 | 1634 > 833, 417, 200, 100 |
| 833, 417 > 100 | |||
| Female | NS | ||
| Both | p < 0.05 | 1634 > 833, 200, 100 | |
| 833, 417, 200 > 100 |
The ANOVAs and post hoc tests indicate a significant increase in scores for the male and combined talkers for SR9, with an increase in the cutoff frequency from 12.5 to 25 Hz (see left column of Fig. 5). Further increases in cutoff frequency do not produce further significant increases in scores. The ANOVA for the female talker is not significant, indicating that manipulations in cutoff frequency over the tested range do not affect scores with that talker for this subject.
The ANOVAs and post hoc tests indicate significant increases in scores for each of the talker conditions for SR9 with increases in rate (see right column of Fig. 5). The highest scores are obtained at or near the highest tested rate for either of the talkers. The largest increases in scores for the male talker are produced with the increase in rate from 50 to 100 pulses/s/channel and from 417 to 1634 pulses/s/channel, whereas the largest increases in scores for the female talker are produced with increases in rate from 417 to 2525 pulses/s/channel. The score at 2525 pulses/s/channel is significantly higher than the score at 1634 pulses/s/channel. Increases in scores for the combined talkers are obtained with an increase in rate from 50 pulses/s/channel to any of the intermediate rates between 100 and 417 pulses/s/channel and from that range of rates to the higher tested rates.
For SR10, none of the ANOVAs for manipulations in cutoff frequency is significant, at least for the range of cutoffs included for all talker conditions (see left column of Fig. 6). This indicates that changes in cutoff between 50 and 800 Hz did not affect scores. However, inclusion of data for lower cutoffs does produce a significant ANOVA for the male talker (p < 0.01; see additional gray symbols and lines in the top left panel of Fig. 6). Post hoc tests for that case indicate that scores for cutoffs at and above 100 Hz are significantly higher than the scores for the cutoffs of 12.5 and 25 Hz, and that the score for the 50 Hz cutoff is higher than the score for the 12.5 Hz cutoff, as previously reported in QPR 6.
The ANOVAs for manipulations in rate indicate significant differences among scores for the male and combined talkers, but not for the female talker, for the range of rates common to all talker conditions. For the male talker, the score at the highest tested rate of 1634 pulses/s/channel is significantly higher than the scores at all lower rates, and the scores at 833 and 417 pulses/s/channel are significantly higher than the score at 100 pulses/s/channel. For the combined talkers, the score at 1634 pulses/s/channel is significantly higher than the scores at 833, 200 and 100 pulses/s/channel (but not at 417 pulses/s/channel), and the scores at 833, 417 and 200 pulses/s/channel are significantly higher than the score at 100 pulses/s/channel.
The pattern of scores for the female talker in the right column of Fig. 6 is similar to the patterns for the male and combined talker conditions. However, the ANOVA for the female talker is not significant, whereas the ANOVAs for the other conditions are significant. A possible contributor to this seemingly curious result is that only one point is included in the "average" of scores for the lowest rate of 100 pulses/s/channel. Thus, the apparently-large decrement in the score at that rate contributes little to the ANOVA, which includes the average scores at each of the higher rates, all computed with higher numbers of scores. For the ANOVA to be significant, there must either be an even larger difference between the score at 100 pulses/s/channel and the scores at all other rates, or there must be large differences among scores at those other (higher) rates. The data for subject SR10 show the latter differences for the male talker and combined talkers, but not for the female talker.
Inclusion of all data for the male talker reduces slightly the differences among scores across rates for SR10 (see gray symbols and lines in the top right panel of Fig. 6). The ANOVA for all data just fails to attain significance (p = 0.083).
CUNY sentences. Results from the tests of sentence recognition are presented in Fig. 7, which shows interpolated and extrapolated contour plots of the percent correct scores for subjects SR9 and SR10. The percent correct scores underlying these plots, along with the standard errors of the means for the scores, are presented in Appendix Fig. A.2. Four separate lists of the CUNY sentences were used in the measures for each processor with each subject. The lists were not repeated within or across conditions for either of the subjects.
The plots in Fig. 7 show contours over all possible combinations of
rate and cutoff frequency, for the indicated ranges of those parameters.
However, combinations with cutoff frequencies higher than one-half the
rate of stimulation were not tested with either subject. Individual points
and their projections to the rate/cutoff frequency plane are provided to
indicate where the contours represent interpolation or extrapolation of
the data.
Fig. 7. Recognition of CUNY sentences by subjects
SR9 and SR10, male talker. Percent correct scores are shown as points;
their projections to the rate/cutoff frequency plane and interpolated and
extrapolated contour plots are provided as well. Matrices of the percent
correct scores and associated standard errors of the means are presented
in Appendix Fig. A.2. Note that the ranges for the rate and cutoff frequency
(LPF Cutoff) axes in this figure are different from those in Fig. 1.
Quantitative comparisons among the scores are presented in Fig. 8, which
shows difference matrices for the two subjects. A one-way ANOVA was conducted
for each matrix and was significant for subject SR9 but not for subject
SR10.
Fig. 8. Difference matrices for recognition of
CUNY sentences, male talker, subjects SR9 and SR10. Each matrix was derived
by subtracting the mean of all percent correct-scores for that matrix (see
Appendix Fig. A.2) from the percent-correct score in each cell. A one-way
analysis of the variance (ANOVA) was conducted to evaluate the possibility
of significant differences among the scores for each matrix. The ANOVA
was significant for SR9 but not for SR10. Shaded areas in the matrix for
SR9 indicate scores that are significantly different from the mean according
to the Fisher LSD criterion (p < 0.05; Fisher LSD = 12.9), and
the bold type indicates the one score that is significantly different from
the mean according to the Tukey HSD criterion (p < 0.05; Tukey
HSD = 23.04). For this matrix all scores significantly different from the
mean are below the mean. Underlines indicate the maximum and minimum values
in the matrix for SR9.
The results from the CUNY tests are similar to the results from the consonant tests for subject SR9. The shape of the contour plot for CUNY sentences (left column of Fig. 7) approximates the shapes of the plots for consonant identification for each of the talker conditions (three plots in the left column of Fig. 1; note that the ranges of rates and cutoff frequencies are both greater in these plots for consonant identification than in the plot for the CUNY sentences). The maximum scores for both the CUNY and consonant tests are obtained with the highest tested rate of stimulation and either an intermediate or the highest tested cutoff frequency. The minimum scores are obtained at the lowest tested cutoff frequency or at the lowest tested rate or both.
For SR10, the contour plot for the CUNY sentences is much flatter than the plots for consonant identification. Indeed, the ANOVA for the CUNY scores is not significant.
In broad terms, results from tests with the CUNY sentences reflect results from tests of consonant identification for subject SR9. The tests with the CUNY sentences appear to be as sensitive, or perhaps somewhat more sensitive, as the consonant tests in demonstrating effects of rate and cutoff frequency manipulations for this subject. In contrast, the CUNY tests are not as sensitive as the consonant tests in demonstrating such effects for subject SR10. Note, however, that range of CUNY scores is quite different between the subjects. Scores for SR9 range from 29 to 68 percent correct, whereas scores for SR10 range from 80 to 98 percent correct. Most of the scores for SR10 are 90 percent correct or higher. Thus, ceiling effects may well have diminished the sensitivity of the CUNY tests for that subject.
Consonants presented in competition with noise. Effects of noise interference were evaluated in tests of consonant identification with subject SR2. The conditions included especially wide ranges of processor parameters and consonants presented in quiet and at the speech-to-noise ratios of +10 and +5 dB. As noted before, CCITT noise was used, which has a spectrum that approximates the long-term spectrum of speech.
The principal results are presented in Figs. 9 and 10, which show contour
plots and difference matrices, respectively, for the quiet and +10 and
+5 dB speech-to-noise conditions. The percent correct scores used to derive
these figures are presented in Appendix Fig. A.3. Figure A.3 also shows
the standard errors of the means for the scores.
Fig. 9. Consonant identification with the male
talker in quiet and in competition with speech-spectrum noise at the speech-to-noise
ratios of +10 and +5 dB, subject SR2. Percent correct scores are shown
as points; their projections to the rate/cutoff frequency plane and interpolated
and extrapolated contour plots are provided as well. Matrices of the percent
correct scores and associated standard errors of the means are presented
in Appendix Fig. A.3. Some of the scores for the speech-to-noise ratio
of +10 dB are for tests conducted under questionable conditions: a possible
flaw in the implementation of one processor and exceptionally strong tinnitus
during the testing of several others. These questionable scores are identified
in Fig. 10. The right panel in the present figure shows a contour plot
derived from the scores that remained after all questionable scores were
removed.
Fig. 10. Difference matrices for tests of consonant
identification in quiet and in competition with speech-spectrum noise at
the speech-to-noise ratios of +10 and +5 dB, male talker, subject SR2.
Each matrix was derived by subtracting the mean of all percent correct-scores
for that matrix (see Appendix Fig. A.3) from the percent-correct score
in each cell. A one-way analysis of the variance (ANOVA) was conducted
to evaluate the possibility of significant differences among the scores
for each matrix. The ANOVAs indicated significant differences for all matrices
in the figure. Some of the scores for the speech-to-noise ratio of +10
dB are for tests conducted under questionable conditions: a possible flaw
in the implementation of one processor (corresponding to the cell marked
with a superscripted "a" in the middle matrix in the left column) and exceptionally
strong tinnitus during the testing of several others (corresponding to
the cells marked with a superscripted "b"). A difference matrix computed
from the scores that remained after the questionable scores were removed
is presented in the right column. The meanings of shadings, bold type,
underlined type, and heavy lines around cells are presented in the caption
to Fig. 2.
As noted in the caption to Fig. 10, some of the scores for the +10 dB speech-to-noise ratio (S/N) were obtained under questionable conditions. The score marked with a superscripted "a" in Fig. 10 was obtained with a processor whose implementation may have been flawed (and was identified at the time as a processor that needed to be checked and tested again). The scores marked with a superscripted "b" in Fig. 10 were obtained during a common period when the subject volunteered that he was experiencing especially severe tinnitus, which may have affected his performance in the tests.
Given the questions associated with some of the scores for the +10 dB conditions, a separate contour plot and a separate difference matrix were derived from a subset of the collected data with the questionable points removed. The additional contour plot is shown in the right column of Fig. 9, and the additional difference matrix is shown in the right column of Fig. 10. The corresponding plot and matrix in the left columns of Figs. 9 and 10 were derived from all of the collected data.
A one-way ANOVA was conducted for each of the matrices in Fig. 10 and
in each case indicated significant differences among the scores. Results
from post hoc tests using the Fisher LSD and Tukey HSD criteria
are indicated by shading and boldface type, as described in the caption
to Fig. 2. Underlines identify the maximum and minimum scores in each matrix.
Critical values for comparisons using either the Fisher LSD or Tukey HSD
are presented in Table 4.
Table 4. Critical values for comparisons between scores within
matrices in Figs. 10 and A.3.
| Fisher LSD | Tukey HSD | |||||
| Noise Condition | outsidea | acrossb | withinc | outsidea | acrossb | withinc |
| Quiet | 6.81 | 5.89 | 4.81 | 13.89 | 12.03 | 9.82 |
| +10 dB S/N, all points | 7.77 | 15.86 | ||||
| +10 dB S/N, questionable points removed | 7.65 | 15.35 | ||||
| +5 dB S/N | 8.02 | 7.32 | 16.38 | 14.95 |
As described in QPR 6, results for the consonants presented in quiet show decrements in performance with low rates and low cutoff frequencies. Conversely, scores that are significantly above the mean of all scores are obtained with relatively high rates and relatively high cutoff frequencies. The very lowest scores are obtained with rates at and below 125 pulses/s/channel and cutoff frequencies at and below 50 Hz. The highest score is obtained at the highest tested rate (6900 pulses/s/channel) and at a relatively high cutoff frequency (500 Hz).
The pattern of scores for consonants presented in competition with noise at the S/N of +10 dB is somewhat similar to the pattern for consonants presented in quiet. The lowest scores are again obtained with low rates and cutoff frequencies, and the highest scores are again obtained with combinations of relatively high rates and relatively high cutoff frequencies. The minimum score for consonants in quiet is at the rate of 50 pulses/s/channel and the cutoff frequency of 25 Hz, and the minimum score for consonants presented in noise at the +10 dB S/N is at that rate and the nearby cutoff frequency of 50 Hz. Similarly, the maximum score for consonants in quiet is at the rate of 6900 pulses/s/channel and the cutoff frequency of 500 Hz, and the maximum score for consonants presented in noise at the +10 dB S/N is at that rate and the nearby cutoff frequency of 250 Hz.
A difference in the patterns of scores for consonants presented in quiet versus consonants presented in noise at the +10 dB S/N is that the dependence on both rate and cutoff frequency appears to be more gradual for the consonants presented in noise. The contour plots for the +10 dB conditions show a more-or-less continuous reduction in scores with reductions in rate across the tested range, and a more-or-lees continuous reduction in scores with reductions in cutoff frequency from 100 Hz to the lowest tested value of 12.5 Hz. In contrast, the plot for consonants presented in quiet shows a more precipitous decrement in performance with combinations of very low rates and low cutoff frequencies, and a smaller region of relatively high scores with combinations of the highest tested rate and intermediate cutoff frequencies.
Comparison of the left and right columns in Figs. 9 and 10 shows that the overall pattern of results for the +10 dB S/N is not affected by removal of the questionable points. Indeed, only subtle differences can be found between the two contour plots presented in Fig. 9 for the +10 dB S/N. (We note that the data points we have identified as having been obtained under questionable conditions may, in fact, reflect a significant additional feature in the performance contour. We plan to revisit those parameter combinations at the next opportunity with the same subject.)
The pattern of scores for consonants presented at the +5 dB S/N is similar to the pattern for consonants presented in quiet. Both patterns show precipitous decrements in performance with combinations of low rates and low cutoff frequencies, and both patterns show better than average performance with relatively high rates and relatively high cutoff frequencies. The patterns for quiet and +5 dB S/N also show broad regions of relatively uniform scores, in contrast to the pattern observed for the +10 dB S/N.
Not surprisingly, the addition of competing noise reduces the mean of all scores for the various processor conditions. The mean of all scores for consonants presented in quiet is 71.8 percent correct, whereas the means for the +10 and +5 dB noise levels are 47.1 and 37.9 percent correct, respectively. (The mean for the +10 dB level with the questionable points removed is 49.5 percent correct.)
Slices through the rate/cutoff frequency data are presented in Figs.
11 and 12. Fig. 11 shows results for the consonants presented in quiet
and at the two speech-to-noise ratios, and Fig. 12 shows results with the
questionable points removed for the +10 dB S/N. Effects of manipulations
in cutoff frequency are shown in the left columns of the two figures, and
effects of manipulations in rate are shown in the right columns.
Fig. 11. Slices through the rate/cutoff frequency
data for subject SR2, from tests of consonant identification using the
male talker and with the consonants presented in quiet and in competition
with speech-spectrum noise at the speech-to-noise ratios of +10 and +5
dB. The left column shows slices along rows in the matrices of the percent
correct scores for this subject (Fig. A.3), and the right column shows
slices along columns in the matrices. Plots in the left column indicate
the dependence of scores on cutoff frequency while holding the rate of
stimulation constant, whereas plots in the right column indicate the dependence
of scores on rate of stimulation while holding the cutoff frequency constant.
Standard errors of the means for the scores ranged from 1.2 to 4.7 percent
and are presented in Fig. A.3. The legend for the left column indicates
rates of stimulation in pulses/s/channel (pps/ch), and the legend in the
right column indicates envelope cutoff frequencies in Hz (LPF). All of
the data collected for the +10 dB speech-to-noise ratio are presented in
this figure. Fig. 12 shows the data for that noise condition with the questionable
points (identified in Fig. 10) removed.
Fig. 12. Panels like those in Fig. 11 for the +10
dB speech-to-noise ratio, but with the questionable points (identified
in Fig. 10) removed. Symbols are the same as those in Fig. 11.
As described in QPR 6 for consonants presented in quiet, increases in cutoff frequency produce increases in scores for all rates stimulation except the lowest rate of 50 pulses/s/channel, at which relatively uniform scores are obtained at the three tested cutoff frequencies (top left panel of Fig. 11). The patterns of the increases in scores with increases in cutoff frequency are similar to each other for rates of stimulation at and above 250 pulses/s/channel. The patterns are different for lower rates. Also, the scores for those lower rates are much lower than the scores for rates at and above 250 pulses/s/channel, for some or all of the tested cutoff frequencies with the lower rates. For the higher rates, scores increase more-or-less uniformly with increases in cutoff frequency up to 50 Hz.
The patterns of scores produced with increases in rate for consonants presented in quiet are relatively uniform for cutoff frequencies at and above 25 Hz (top right panel of Fig. 11). Scores for those cutoff frequencies appear to increase with increases in rate from the lowest tested rate of 50 pulses/s/channel up to 125 pulses/s/channel or higher. Scores for the lowest tested cutoff frequency of 12.5 Hz increase with increases in rate up to 250 pulses/s/channel, but are much lower than the scores for all other cutoff frequencies at rates of 100 pulses/s/channel and higher.
Results for consonants presented in conjunction with noise at the +10 dB S/N are more variable than the results for consonants presented in quiet (compare top and middle rows of Fig. 11). The patterns of scores produced with manipulations in cutoff frequency for the +10 dB S/N do not cluster together for rates at and above 250 pulses/s/channel, as observed for consonants presented in quiet. Instead, most scores increase with progressively higher rates over the range of cutoff frequencies from about 125 Hz to the upper limit of 2000 Hz. The patterns also suggest effects of changes in cutoff frequencies over certain ranges. Scores are approximately the same between the two lowest tested cutoff frequencies of 12.5 and 25 Hz, for most of the rates. Scores increase with increases in cutoff frequency from 25 to 125 Hz, for rates at and above 1000 pulses/s/channel. This general increase with increases in cutoff is more apparent with the questionable points removed (compare the middle left panel of Fig. 11 with the left panel of Fig. 12). Scores are approximately the same across cutoff frequencies from 125 to 1000 Hz, for rates at and above 2000 pulses/s/channel. With the further increase in cutoff frequency, from 1000 to 2000 Hz, scores appear to drop.
With the questionable points removed, the 50 pulses/s/channel rate clearly produces the lowest scores over the range of cutoff frequencies tested at that rate. In addition, the variability among scores, across cutoff frequencies and for the different rates, appears to be somewhat lower in the panel with the questionable points removed (left panel of Fig. 12), compared with the panel presenting all of the collected data (left middle panel of Fig. 11). The impression of a drop in scores with increases in cutoff frequency beyond 500 or 1000 Hz is stronger in the panel with all of the collected data than in the panel with the questionable points removed.
Increases in rate appear to produce increases in scores over wide ranges of the tested rates for consonants presented at +10 dB S/N (right middle panel of Fig. 11). This is most evident for the cutoff frequencies of 250 to 1000 Hz, for which monotonic and generally large increases in scores are observed across the ranges of cutoff frequencies tested with those rates. Scores for the two lowest cutoff frequencies show a different pattern with the manipulations in rate. They increase with the increase in rate from 50 to 125 pulses/s/channel, but then vary together below and up to the level obtained at 125 pulse/s/channel, with further increases in rate. This observation holds with and without the questionable points removed.
The patterns of scores for consonants presented with noise at the +5 dB S/N appear to be less variable than the patterns at the +10 dB S/N (compare bottom and middle rows in Fig. 11). However, the overall dependence of scores on changes in cutoff frequency appears to be similar for the two speech-to-noise ratios. In both cases, the increase in cutoff frequency from 12.5 to 25 Hz does not seem to affect scores for most of the tested rates. Increases in cutoff frequency from 25 to 100 Hz appear to produce increases in scores at both S/Ns, for most or all rates at and above 1000 pulses/s/channel. Further increases in cutoff frequency up to 1000 Hz do not produce changes in scores for either S/N, with the possible exception of a decrement in scores with increases in cutoff frequency for the 1000 pulses/s/channel rate tested at the +10 dB S/N. The final increase cutoff frequency from 1000 Hz to 2000 Hz appears to produce a drop in scores for both sets of data.
Scores for the various cutoff frequencies tested at the 50 pulses/s/channel rate are much lower than all other scores for the +5 dB S/N. This finding is consistent with the findings for consonants presented in quiet and at the +10 dB S/N.
Increases in rate appear to produce increases in scores over much or all of the tested range of rates for the +5 dB S/N (see bottom right panel of Fig. 11). Scores for the different cutoff frequencies vary together with the changes in rate. Scores across rates for the lowest tested cutoff frequency of 12.5 Hz are not significantly lower than the scores for higher cutoff frequencies, except at the extreme rates of 50 and 6900 pulses/s/channel.
The averages of scores in Figs. 11 and 12 are shown in Fig. 13. The
left column of Fig. 13 presents averages for all tested rates of stimulation
above 200 pulses/s/channel (as in Figs. 5 and 6) versus cutoff frequency.
The right column of Fig. 13 presents averages for all tested cutoff frequencies
at and above 25 Hz (also as in Figs. 5 and 6) versus rate of stimulation.
The gray symbols and lines in the middle panels of Fig. 13 show the averages
computed with the questionable scores removed for the +10 dB S/N case.
Fig. 13. Averages of scores derived from data in
Figs. 11 and 12., for subject SR2. The plots in the left column of the
present figure show the averages of scores for all tested rates of stimulation
above 200 pulses/s/channel (i.e., 250 pulses/s/channel and higher for this
subject), and the plots in the right column show the averages of scores
for all tested cutoff frequencies at and above 25 Hz. The gray symbols
and lines show results with the questionable points removed (for the +10
dB speech-to-noise ratio). The numbers in the bottom panels for each column
indicate the number of scores included in calculations of the means for
each cutoff frequency (left column) or each rate (right column). The gray
numbers in the middle panels indicate those numbers for the calculations
with the questionable points removed. The bars show standard errors of
the means.
One-way ANOVAs were conducted for each set of averages in Fig. 13. The
results are presented in Table 5. All of the ANOVAs were significant except
the one for the averages across cutoff frequencies for the +10 dB S/N with
all points included (middle left panel of Fig. 13, black symbols and lines).
That ANOVA just failed to attain significance (p = 0.073). (The
variabilities associated with each average are higher with all points included,
and the differences among the averages are lower with all points included,
each of which reduces the F value for the ANOVA.)
Table 5. Significant differences among scores for the cutoff
frequency and rate of stimulation conditions presented in Fig. 13 (data
for subject SR2). The differences indicated are those identified by the
Fisher LSD criterion (p < 0.05). Entries for cutoff frequencies
are in Hz, and entries for rate of stimulation are in pulses/s/channel.
| Noise condition | Manipulation | ANOVA | Significant Differences |
| Quiet | Cutoff frequency | p < 0.001 | 2000, 1000, 500, 250, 125, 50 > 25, 12.5 |
| 25 > 12.5 | |||
| +10 dB S/N, all points | NS | ||
| +10 dB S/N, questionable points removed | p < 0.001 | 1000, 500, 250, 125 > 50, 25, 12.5 | |
| +5 dB S/N | p < 0.01 | 125 > 2000, 25, 12.5 | |
| 1000, 500, 250 > 25, 12.5 | |||
| Quiet | Rate | p < 0.001 | 6900, 3500, 2000, 1000, 500 > 125, 50 |
| 250, 125 > 50 | |||
| +10 dB S/N, all points | p < 0.001 | 6900, 3500, 2000 > 500, 250, 125, 50 | |
| 1000 > 50 | |||
| +10 dB S/N, questionable points removed | p < 0.01 | 6900, 3500, 2000, 1000 > 50 | |
| +5 dB S/N | p < 0.001 | 6900 > 500, 125, 50 | |
| 3500, 2000, 1000 > 125, 50 | |||
| 500, 250, 125 > 50 |
Results from post hoc tests (also presented in Table 5) identify effects of manipulations in both cutoff frequency (left column of Fig. 13) and rate of stimulation (right column of Fig. 13). Effects of manipulations in cutoff frequency include significant increases in scores with increases in cutoff frequency up to 50 Hz for consonants presented in quiet. The score at 25 Hz is significantly higher than the score at 12.5 Hz, and the score at 50 Hz is significantly higher than the score at 25 Hz. Further increases in cutoff frequency do not produce further significant increases in scores.
The addition of noise interference alters the pattern of averages across cutoff frequencies. For the both the +10 and +5 dB S/Ns, averages are not significantly different between the 12.5 and 25 Hz cutoffs. Averages for cutoffs between 125 and 1000 Hz are, however, significantly higher than the averages for lower cutoffs. For the +10 dB S/N with the questionable points removed, averages for cutoffs in the range from 125 to 1000 Hz are significantly higher than averages for each cutoff up to 50 Hz, and for the +5 dB S/N, averages for cutoffs in that upper range are significantly higher than averages for each cutoff up to 25 Hz.
A further increase in cutoff frequency, from 1000 to 2000 Hz, produces a significant drop in averages for the +5 dB S/N. In particular, the average at the 125 Hz cutoff is significantly higher than the average at 2000 Hz. (The similar drop at 2000 Hz for the +10 dB S/N is not significant.)
The patterns of averages across rates of stimulation also are affected by noise interference. In quiet, averages at 500 pulses/s/channel and above are significantly higher than the averages at 50 and 125 pulses/s/channel. Also, the averages at 125 and 250 pulses/s/channel are higher than the average at 50 pulses/s/channel. The highest rate beyond which further increases in averages are not produced with further increases in rate is 500 pulses/s/channel.
The patterns for consonants presented in noise show advantages of higher rates of stimulation. For the +10 dB S/N with all points included, averages at 2000 pulses/s/channel and above are significantly higher than the averages for all rates from 50 to 500 pulses/s/channel. In addition, the average at 1000 pulses/s/channel is higher than the average at 50 pulses/s/channel.
For the +10 dB S/N with the questionable points removed, averages at 1000 pulses/s/channel and above are significantly higher than the average at 50 pulses/s/channel. For the +5 dB S/N, the average at the highest tested rate of 6900 pulses/s/channel is significantly higher than the averages at 50, 125 and 500 pulses/s/channel. Also, the averages for rates from 1000 to 3500 pulses/s/channel are higher than the averages at 50 and 125 Hz, and the averages for rates from 125 to 500 pulses/s/channel are higher than the average at 50 pulses/s/channel. In all, the results demonstrate improvements in the averages with increases in rate up to 500 pulses/s/channel for consonants presented in quiet, up to 1000 or 2000 pulses/s/channel for consonants presented at the +10 dB S/N, and up to 1000 pulses/s/channel for consonants presented at the +5 dB S/N. The further increment in averages at 6900 pulses/s/channel for the +5 dB S/N suggests the possibility that even higher rates may have been helpful there.
Discussion
The present study extended a prior study to evaluate effects of changes in stimulus rate and envelope cutoff frequency on the speech reception performance of CIS processors. The present study included (1) tests of consonant identification using a female talker, for two subjects previously tested with a male talker; (2) tests of sentence recognition for those same two subjects; and (3) tests of consonant identification in noise using the male talker, for another of the previously-tested subjects. Many combinations of rates and cutoff frequencies were included in each set of tests.
Results from the tests with the female talker were similar to the prior results with the male talker. The contour plots of percent correct scores for the female talker had the same general shape as the plots for the male talker, for both subjects. Combinations of low rates and low cutoff frequencies produced large decrements in scores for both talkers and both subjects. Conversely, the highest scores for both talkers and both subjects were obtained at the highest tested rates and at intermediate cutoff frequencies. Patterns of scores for the combined male and female talkers were only somewhat different from the patterns for the male talker, described before in QPR 6.
Although some differences were noted in the patterns of scores between the male and female talkers for each of the subjects, the results did not provide evidence of a consistent talker effect for the conditions of the present and prior studies. The present results with the female talker confirm and extend the prior results with the male talker.
Results from the tests of sentence recognition were consistent with the results from the consonants tests for subject SR9 but not for subject SR10. Sentence scores for SR9 ranged from 29 to 68 percent correct across the various combinations of rate and cutoff frequency, whereas the scores for SR10 ranged from 80 to 98 percent correct, with most of those scores at 90 or higher. Ceiling effects may have limited the sensitivity of the sentence test for SR10. If so, this would explain the differences in the patterns of consonant and sentence scores for him.
The results from the sentence tests verify and extend the prior results (from the consonant tests with the male talker) for subject SR9. In contrast to the results from the consonant tests, scores from the sentence tests for subject SR10 are not significantly different across the various combinations of rate and cutoff frequency. This lack of significance may be attributable to ceiling effects in the sentence tests for this subject.
Patterns of scores from the final set of tests showed that both higher cutoff frequencies and higher rates are advantageous for consonant identification in noise, compared with consonant identification in quiet. In quiet, scores improved with increases in rate up to 500 pulses/s/channel and with increases in cutoff frequency up to 50 Hz. At the S/N of +10 dB, these points of asymptotic performance were 1000 or 2000 pulses/s/channel and 125 Hz, respectively (the two numbers for rate correspond to results from analyses with and without the questionable points removed). At the S/N of +5 dB, the points were 1000 pulses/s/channel and 125 Hz, with a suggestion that even higher rates (at or beyond the upper limit of the tested rates) might be helpful. At that S/N, a decrement in scores also was observed with the increase in cutoff frequency from 1000 to 2000 Hz (the upper limit of tested cutoffs). Possibly, an especially high cutoff frequency may degrade performance in noise, in that such a cutoff might allow interference from high-frequency and random variations in the noise, whereas lower cutoffs might reduce or eliminate the influence of high-frequency variations.
One of the major problems associated with present implant systems is
the great difficulty their users experience in listening to speech in noise.
The present results suggest that this difficulty may be substantially reduced
through the use of relatively high rates and relatively high cutoff frequencies
in the speech processors used with these systems.
III. Plans for the next quarter
Our plans for the next quarter include the following:
We thank subjects SR3, ME3, SR1 and SR8 for their participation in the
studies of this quarter. We also would like to thank again subjects SR2,
SR9 and SR10 for their participation in the studies of prior quarters,
as described in section II of this report. We are grateful for the many
contributions made by Raymond Mederake
in the studies with subject ME3.
V. Erratum for Quarterly Progress Report 6
Table 1 in QPR 6 for this project did not include a reference to Lawson
et
al., 1993 (Quarterly Progress Report 2, NIH project N01-DC-2-2410).
This study predated all studies that were included in the original table.
A revised Table 1 is presented below. The results presented in Lawson et
al., 1993, do not alter the conclusions of QPR 6.
| Study | Subjects | Channels | Rate(s) | LPF Cutoff(s) | Test(s) |
| Brill et al., 1997 | 2 | 2
3 4 6 8 10 |
1515, 9090
1515, 6060 1515, 4545 1515, 3030 1515, 2273 1515, 1818 |
400 | HSM sent. w. lipreading, subj. MK; 2-digit numbers, subj. FZ |
| Brill et al., 1998 | 1
5 4 |
8
6 4 |
625, 694, 781, 893, 1042, 1250, 1515
400, 800, 1515, 3030 800, 1515, 3030, 4545 |
400 for rates of 1515 and higher; ¼ the rate for lower rates | 16 consonants
monosyllabic words monosyllabic words or 2-digit numbers |
| Fu & Shannon, 2000 | 6 | 4 | 50, 100, 150, 200, 300, 400, 500
500 |
40 % of the rate
2, 5, 10, 20, 40, 80, 160 |
12 vowels, 16 consonants |
| Kiefer et al., 2000 | 12
6 |
4
5 or 6 7 or 8 7 or 8 4 |
600, 1200, 1515-1730, max rate
1515-1730, max rate 600, 1200, 1515-1730 600 max rate |
400
200 26.67 % of the rate |
8 vowels, 16 consonants, monosyllabic words |
| Lawson et al., 1993 | 1 | 4 | 417
627 2525 |
50, 200, 800
50, 200 50, 200, 800 |
24 consonants |
| Lawson et al., 1996 | 5 | 6 | 250, 833, 2525
833 |
200
400 |
16 or 24 consonants, dep. on subj. |
| Loizou & Poroy, 1999 | 6 | 6 | 400, 800, 1400, 2100 | 100, 200, 400 and 400, respectively | 11 vowels, 20 consonants, monosyllabic words |
| Pelizzone et al., 1998 | 4 | 5 | 2000, 20000 | 400 | 7 vowels, 14 consonants |
Table 1. Prior and contemporaneous studies on effects of changes
in stimulus rate and envelope cutoff frequency for CIS processors.
Appendix 1. Summary of reporting activity for this quarter
Reporting activity for this quarter, covering the period of April 1 through June 30, 2000, included the following presentations:
Brill S: Electrode discrimination along the cochlea based on pitch perception. Invited talk presented at the Biweekly Meeting of the Duke/RTI Cochlear Implant Team, Duke University Medical Center, Durham, NC, May 28, 2000.
Wilson BS, Lawson DT, Brill S, Wolford RD and Schatzer R: Binaural cochlear implants. Invited lecture presented at the Conference on Binaural Hearing, Hearing Loss, Hearing Aids, & Cochlear Implants, Iowa City, IA, June 22-24, 2000.
Tyler R, Parkinson A, Gantz B, Rubinstein J, Wilson B, Witt S, Wolaver
A, Lowder M: Independent binaural cochlear implants. Invited
lecture presented at the Conference on Binaural Hearing, Hearing Loss,
Hearing Aids, & Cochlear Implants, Iowa City, IA, June 22-24, 2000.
Appendix
2. Supplemental data for Section II
Fig. A.1. Percent correct scores and standard errors
of the means for identification of consonants in quiet, male and female
talkers, subjects SR9 and SR10.
Fig. A.2. Percent correct scores and standard errors
of the means for recognition of CUNY sentences in quiet, male talker, subjects
SR9 and SR10.
Fig. A.3. Percent correct scores and standard errors
of the means for identification of consonants in quiet and in competition
with speech-spectrum noise at the speech-to-noise ratios of +10 and +5
dB, male talker, subject SR2. Some of the scores for the tests at the speech-to-noise
ratio of +10 dB are questionable and these scores (and associated processor
conditions) are identified in Fig. 10.