Results
In total, 94 children (47 girls and 47 boys) with hearing loss were
recruited, and their ages ranged from
30
to 180 months old (mean 79.6±37.3 months, median 72 months). All 94
children completed the test. In brief, SIS1 were the
highest, followed by SIS2 and then
SIS3. The interquartile range (IQR) for
SISASR was 40 to 92, with a median of 72. The IQR for
SIS1 was 77.5 to 100, with a median of 96. The IQR for
SIS2 was 74 to 100, with a median of 94. The IQR for
SIS3 was 71.4 to 98.7, with a median of 89. The scores
of the tests are shown in Table 1.
To intuitively analyze the differences between different raters, 20 of
the 94 children were randomly sampled in the final analysis using the
SPSS software program. Among them, the SISASR for 17
children was lower than the subjective assessments. Only the
SISASR for two children was higher than the
SIS1. Only the SISASR for one child
was higher than the SIS3. Figure 1 shows the scores of
subjective evaluations (including three levels) and the scores of the
automatic speech recognition system.
The correlations between the SISASR and the
SISTE are shown in Figure 2. The overall trend showed
that the higher the SISASR, the higher the
SISTE, indicating a positive correlation between the
scores.
SISASR showed perfect consistency with the
SISTE, as shown in Table 2. The lowest correlation
between a rater and the mean of the other two levels was 0.928
(SIS1 vs SIS3), and the highest was
0.971 (SIS2 vs SIS3). Table 2 also
shows the correlations between different raters. The correlation between
the traditional evaluation group and automatic speech recognition is
0.883. This means that the agreement between the human raters and the
machine and the agreement among the humans alone can be regarded as
significant.