|
|
Preface
|
|
|
|
|
|
B. Granstrom. The latter two researchers devoted a year to this project on leave
|
|
|
from the Royal Institute of Technology in Stockholm, Sweden. In addition to the
|
|
|
durational framework, a comprehensive investigation of fundamental frequency ef-
|
|
|
fects was made by J. Allen, D. O’Shaughnessy, and A. Weibel. O’Shaughnessy’s
|
|
|
doctoral thesis contains an extensive compendium of these results, and he is
|
|
|
responsible for the fundamental frequency generation algorithm currently imple-
|
|
|
mented in MITalk. A. Weibel contributed a characterization of fundamental fre-
|
|
|
quency contours in questions.
|
|
|
|
|
|
Given the prosodic framework mentioned above, phonetic segment labels,
|
|
|
stress marks, and junctural marks provided by the syntactic analysis, then
|
|
|
phonemic synthesis routines can be utilized to produce the output speech
|
|
|
waveform. The MITalk system is based on a phonemic speech synthesis model
|
|
|
developed by D. H. Klatt. All of the algorithms for the specification of the control
|
|
|
parameters utilized by this model were developed by him. During the stay of
|
|
|
R. Carlson and B. Granstrom, further refinements, modifications, and tests were
|
|
|
performed in the context of the overall MITalk system. At that time, many issues
|
|
|
concerned with consistency and the integration of the entire system were ad-
|
|
|
dressed.
|
|
|
|
|
|
In the late 1970s, the computational environment for the research was
|
|
|
changed from the PDP-9 computer to a DEC-System 20, with output speech
|
|
|
provided by a PDP-11. A special interface was constructed between the DEC-20
|
|
|
and the PDP-11, and an all-digital special purpose speech synthesis processor was
|
|
|
constructed by G. S. Miranker. This processor was capable of exercising the
|
|
|
phonemic synthesis model in real-time. The DEC-System 20, a large time-shared
|
|
|
machine, was ideally suited to the modular nature of the MITalk system. It per-
|
|
|
mitted many researchers individually and interactively to build the system’s over-
|
|
|
all structure. Beginning in the mid ’70s, a great deal of attention was focused on
|
|
|
the MITalk system’s overall organization. The problems of coordinating such a
|
|
|
large system with its many contributors cannot be overemphasized. As a result,
|
|
|
standard interfaces were established between all modules. Over the years, ex-
|
|
|
tremely valuable system programming contributions were provided by
|
|
|
E. R. Jensen, F. X. Carroll, R. S. Goldhor, G. E. Kopec, and Y. Willems.
|
|
|
|
|
|
As the entire system was built in a coordinated manner, and as experience
|
|
|
with the interaction of all constituent algorithms increased, there was a clear neces-
|
|
|
sity for a comprehensive evaluation of the system. Fortunately, D. Pisoni visited
|
|
|
the Research Laboratory of Electronics and was attracted to the problem of percep-
|
|
|
tual evaluation. He performed a broad review of the testing literature, extended
|
