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.
39 lines
2.2 KiB
39 lines
2.2 KiB
13
|
|
|
|
Some measures of intelligibility and comprehensionl
|
|
|
|
13.1 Overview
|
|
|
|
As the ten year effort to build an unrestricted text-to-speech system at MIT drew to
|
|
a close, it seemed appropriate to conduct a preliminary evaluation of the quality of
|
|
the speech output with a relatively large group of naive listeners. The results of
|
|
such an evaluation would no doubt prove useful in first establishing a benchmark
|
|
level of performance for comparative purposes, as well as uncovering any
|
|
problems in the current version of the system that might not have been detected
|
|
earlier. In addition to obtaining measures of intelligibility of the speech output
|
|
produced by the text-to-speech system, we were also interested in finding out how
|
|
well naive listeners could comprehend continuous text produced by the system.
|
|
This was thought to be an important aspect of the evaluation of the text-to-speech
|
|
system as a whole, since a version of the current system might eventually be im-
|
|
plemented as a device used for computer-aided instruction or as a functional
|
|
reading machine for the blind (Allen, 1973). Both of these applications are now
|
|
well within the realm of the available technology (Allen et al., 1979).
|
|
|
|
In carrying out the evaluation of the system, we patterned several aspects of
|
|
the testing after earlier work already completed on the evaluation of the Haskins
|
|
Laboratories reading machine project so that some initial comparisons could be
|
|
drawn between the two systems (Nye and Gaitenby, 1973, 1974). However, we
|
|
also added several other tests to the evaluation to gain additional information about
|
|
word recognition in normal sentential contexts and listening comprehension for a
|
|
relatively wide range of narrative passages of continuous text. Data were also col-
|
|
lected on reading comprehension for the same set of materials to permit direct
|
|
comparison between the two input modalities. Traditional measures of listening or
|
|
reading comprehension have not typically been obtained in previous evaluations of
|
|
the quality of synthetic speech output, and therefore, we felt that some preliminary
|
|
data would be quite useful before the major components of the present system
|
|
were implemented as a workable text-to-speech device in an applied context.
|
|
|
|
IThis chapter was written by D. Pisoni in 1978-9.
|
|
|
|
151
|