|
|
Preface
|
|
|
|
|
|
The MITalk system described in this book is the result of a long effort, stretching
|
|
|
from the early 1960s to the present. In this preface, a view is given of the work’s
|
|
|
historical evolution. Within this description, acknowledgements are made of the
|
|
|
project’s many contributions. In recognizing these contributions, it is best to or-
|
|
|
ganize them into four groups. First, there is the development of the MITalk system
|
|
|
itself, its evolution, and the many diverse contributions made to its structure and
|
|
|
content. Second, there was the 1979 summer course which resulted in a com-
|
|
|
prehensive summary of the work to that date, and also provided the occasion to
|
|
|
write a set of course notes. Next, there have been continuing efforts (since 1980)
|
|
|
which included re-writes of the system’s software, and the efforts to organize this
|
|
|
book which involved substantial new writing and rule formulations, and explicit
|
|
|
examples directly keyed to the current working system. Finally, there is the spon-
|
|
|
sorship of the program’s many facets over the years.
|
|
|
|
|
|
In the early 1960s, much interest in speech synthesis emerged within the Cog-
|
|
|
nitive Information Processing Group at MIT’s Research Laboratory of Electronics.
|
|
|
This group, led by M. Eden and S. J. Mason, focused on the development of sen-
|
|
|
sory aids for the blind. Many approaches were taken, but it was recognized that
|
|
|
the development of a reading machine for the blind that could scan printed text and
|
|
|
produce spoken output was a major goal. Research efforts in both character recog-
|
|
|
nition and speech synthesis from text were initiated. By 1968, a functional reading
|
|
|
machine was demonstrated. Once the characters were recognized (using a contour
|
|
|
scanning algorithm), text-to-speech conversion was accomplished in two phases.
|
|
|
First, a morph decomposition analysis of all words was performed by using tech-
|
|
|
niques developed by F. F. Lee (in his 1965 doctoral thesis). A morph lexicon suf-
|
|
|
ficient for these demonstrations was developed. It was anticipated that any excep-
|
|
|
tional words not analyzed into morphs would be pronounced by using spelled
|
|
|
speech. As a result, these words were heard as a sequence of individually
|
|
|
pronounced letters. The dictionary provided names of the phonetic segments for
|
|
|
each morph, and synthesis was performed using the algorithms developed and
|
|
|
published by Holrhes, Mattingly, and Shearme. An analog synthesizer was used to
|
|
|
|
|
|
amplifiers. The demonstration of this system was impressive, although the
|
|
|
|
|
|
1
|