SPEECH SYNTHESIS FOR PHONETIC AND
PHONOLOGICAL MODELS
IGNATIUS G. MATTINGLY
1. INTRODUCTION
The linguist today, and more especially the phonologist, is aware of
two very active areas of investigation. One of them, generative
phonology, is at the very center of his field of vision, and cannot
be ignored. The other, experimental phonetics, may seem less directly
relevant to his concerns, even though he accepts the truism that
phonology rests ultimately on a phonetic basis. As modern
experimental phonetics grows more technical, and is increasingly
dominated by the psychologist, the physiologist, the electronics
engineer and the speech scientist, the phonologist is more likely
than ever to be put off and to yield to the temptation to do
phonology on the basis of phonetic folklore.
In this situation, it is fortunate that there is a group of
investigators whose efforts tend to bridge this gap: those
engaged in synthesis of speech by rule. By synthesis by rule,
we mean the automatic production of audible synthetic speech
from a symbolic transcription, by a process which models phonetic
and phonological rules in some nontrivial way. With the help of
modern electronic technology, it is now perfectly possible to type
a transcription on a computer typewriter and immediately hear the
corresponding utterance. But the investigator who undertakes
synthesis by rule not only makes use of advanced technical
facilities, he also attempts of necessity to integrate and
generalize into a system those findings of experimental phonetics
which are relevant to phonology. On this account, at least,
synthesis by rule should intrigue the phonologist.
We believe that there are other reasons as well why the phonologist
should be interested, and we shall try to make them clear below.
We shall also say a little about the techniques of speech synthesis,
give some account of the development of synthesis by rule, describe
a number of current approaches to the task, and finally, suggest
some possible directions which this work may take in the future.
Speech synthesized by rule is not the only kind of synthetic
speech. 1
There are several others which should be mentioned, if only because
investigations motivated by these uses have led to technical
advances of general benefit. Thus, much
__________
1. For general discussions of speech synthesis, see Wheatstone
1837; Dudley and Tarnoczy 1950; Fant 1958; Cooper 1962; and
Flanagan 1965: 167-191.
|