You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
50 lines
2.7 KiB
50 lines
2.7 KiB
Some measures of intelligibility and comprehension
|
|
|
|
B77 tape recorder at 7.5 ips with a 3.0 second pause between successive items.
|
|
Approximately half of the items in a given test list differed in the initial consonant
|
|
|
|
while the remaining half differed in the final consonant.
|
|
|
|
13.2.1.3 Procedure The seventy-two subjects were divided up into twelve inde-
|
|
pendent groups containing six subjects each for testing. Two groups of subjects
|
|
were assigned to each of the six original test lists. Subjects were told that this was
|
|
a test dealing with isolated word recognition and that they were to indicate which
|
|
word out of six possible alternatives was the one they heard on each trial. Forced-
|
|
choice response forms were provided to subjects to record their judgements. Sub-
|
|
jects were encouraged to guess if they were not sure, but to respond on each trial.
|
|
No feedback was provided to subjects during the course of testing. Subjects were,
|
|
however, explicitly informed that the test items were generated on a computer and
|
|
that the experiment was designed to evaluate the intelligibility of the synthetic
|
|
speech. An example of the test format is provided in Appendix D.
|
|
|
|
Testing was carried out in a small experimental room in the Speech Percep-
|
|
tion Laboratory in the Department of Psychology at Indiana University. This room
|
|
is equipped with six individual cubicles. The audio tapes were reproduced on an
|
|
Ampex AG-500 tape recorder and presented to subjects via TDH-39 matched and
|
|
calibrated headphones at a comfortable listening level of about 80 dB SPL peak
|
|
reading on a VTVM. A low-level (60 dB), broad-band (0-10 kHz) white noise
|
|
source (Grason Stadler Model 1724) was also mixed with the speech to mask tape
|
|
hiss, some nonstationary computer-generated background noise picked up during
|
|
the recording at MIT, and any ambient noise in the local environment during test-
|
|
ing.
|
|
|
|
13.2.2 Results and discussion
|
|
|
|
Although the Modified Rhyme Test employed real words, our interest was focused
|
|
on the phoneme errors and resulting perceptual confusions. Overall performance
|
|
on the test was very good with a total error rate, averaged across both initial- and
|
|
final-syllable positions, of only 6.9 percent. Performance was somewhat better for
|
|
consonants in initial position (4.6 percent errors) than final position (9.3 percent
|
|
|
|
€ITOrS).
|
|
The distribution of all errors across various manner classes is shown graphi-
|
|
|
|
calfy in Figure 13-1 for initial- and final-syllable positions separately.
|
|
|
|
Since the consonants comprising the various manner classes occurred with
|
|
unequal frequencies in the Modified Rhyme Test, the observed error rates in the
|
|
data may not be representative estimates of the intelligibility of the same
|
|
phonemes in continuous speech. Nevertheless, performance is generally excellent
|
|
|
|
153
|