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.
45 lines
2.8 KiB
45 lines
2.8 KiB
From text to speech: The MITalk system
|
|
|
|
mance in the second half of the test for subjects in the listening group is consistent
|
|
with our earlier observations in the word-recognition tests which show that listen-
|
|
ing performance improves for synthetic speech after only a short period of ex-
|
|
posure. When the two comprehension groups are compared on the same passages
|
|
in the last half of the test, their performance is equivalent (p > .05), which suggests
|
|
that the overall difference between the two groups is probably due to familiarity
|
|
with the output of the synthesizer and not due to any inherent difference in the
|
|
basic strategies used in comprehending or understanding the content of these pas-
|
|
sages. This conclusion is strengthened even further by the fact that the thirteen
|
|
passages are correlated across both testing conditions. In this case, a very high
|
|
correlation (r = +.97) was observed between reading and listening comprehension
|
|
scores for individual passages. Passages that are difficult to comprehend when
|
|
read are also difficult to comprehend when listened to, and vice versa. The time
|
|
taken to complete all passages in both tests was, however, roughly the same, last-
|
|
ing between 45 and 50 minutes.
|
|
|
|
After the listening comprehension test was completed, we solicited additional
|
|
subjective evaluations of the speech produced by the synthesizer and the nature of
|
|
the comprehension test itself. Twenty of the twenty-two subjects indicated that
|
|
they were able to comprehend and understand the content of the passages “well” or
|
|
“very well”. Only two of the subjects reported difficulty in comprehension, and
|
|
even these two did not indicate that they were merely guessing, an available
|
|
response alternative.
|
|
|
|
Several of the subjects reported improved ability to understand the speech as
|
|
testing progressed. Others described several problems in the quality of synthesis,
|
|
the location of pauses, the existence of inappropriately stressed words, and the oc-
|
|
casional presence of very long “run-on” sentences in several passages. Finally,
|
|
several other subjects suggested that each test passage should be presented twice
|
|
so they could review some of the specific details and facts that were stated ex-
|
|
plicitly. For the most part, however, the subjects found listening to the speech in-
|
|
teresting and felt that they had performed reasonably well in comprehending the
|
|
passages. None of the subjects reported any major distractions in the quality of the
|
|
synthetic speech that interfered with their ability to attend to or understand the con-
|
|
tent of the passages. Thus, subjects are able to adapt easily to relatively long pas-
|
|
sages of synthetic speech with little exposure or practice.
|
|
|
|
13.4.3 Conclusions
|
|
The results of the comprehension test indicate that naive subjects are able to com-
|
|
prehend synthetically produced spoken passages of narrative text output from an
|
|
|
|
166
|