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.
46 lines
2.8 KiB
46 lines
2.8 KiB
Some measures of intelligibility and comprehension
|
|
|
|
unrestricted text-to-speech system. Their performance is roughly comparable to
|
|
subjects who have been asked to read the same passages of text and answer the
|
|
same questions. As in the case of our other tests using synthetic speech, there ap-
|
|
pears to be an initial period during which subjects are simply becoming familiar
|
|
with the quality of the synthesizer, the prosodic rules of the system and the style of
|
|
the material. Even after only a few minutes of exposure, comprehension perfor-
|
|
mance improves substantially and eventually approximates levels observed when
|
|
subjects read the same passages of text.
|
|
|
|
It should also be pointed out that the comprehension performance observed in
|
|
these tests was obtained with a reading rate in excess of 180 words per minute.
|
|
This rate is about the rate at which people typically speak in normal conversations
|
|
or when they read text aloud. The present results therefore suggest that it is not
|
|
necessary to slow down the speaking rate or adjust the synthesis to obtain rela-
|
|
tively high levels of listening comprehension for continuous text. Until the present
|
|
tests were carried out, it was assumed by some investigators that synthetic speech
|
|
had to be output at a much slower rate to maintain intelligibility and therefore
|
|
facilitate comprehension.
|
|
|
|
Based on the results of the present comprehension test, as well as the other
|
|
tests of intelligibility and word recognition that were carried out, there is good
|
|
reason to believe that the basic design of the MIT text-to-speech system is valid.
|
|
The system can not only produce highly intelligible synthetic speech, as shown in
|
|
our earlier tests, but the quality of the synthetic speech can be understood and com-
|
|
prehended at reasonably high levels. While there are, no doubt, many subtle
|
|
details of the system that might be improved, the results of these preliminary tests
|
|
support the general conclusion that very high-quality synthetic speech can be
|
|
produced automatically from unrestricted text and that such a system could be im-
|
|
plemented in applied settings in the immediate future. After some thirty years of
|
|
research, the widespread use of text-to-speech and voice response systems in com-
|
|
puter aided instruction and as aids for the handicapped is now a realistic goal. The
|
|
obstacles are no longer questions of research into the basic principles of speech
|
|
production, perception, and linguistic analysis, but are simply the practical matters
|
|
of implementation and economics.
|
|
|
|
13.5 General discussion and conclusions
|
|
|
|
The results of the three tests designed to evaluate intelligibility, word recognition,
|
|
and listening comprehension indicated very high levels of performance for the cur-
|
|
rent version of the text-to-speech system. While these tests are only preliminary,
|
|
they have provided an initial benchmark against which to compare the perfor-
|
|
|
|
167
|