A programming system for studies in speech synthesis

This paper describes a speech synthesis system which is particularly suitable for experimental investigations. The synthesis is accomplished in two stages. The concatenation stage generates a schematized spectrographic representation corresponding to the symbolic input. The second stage consists in generating the corresponding acoustic signal. The steady state characterization of each phoneme is supplied as data. Independent concatenation procedures incorporate context dependent effects such as format transitions, changes in the normal duration of vowels, etc. The parameter values for these procedures are obtained by a set of rules. Applicability of a rule is determined by attributes assigned to the phonemes. The phonemes are divided into classes and subclasses by the attribute assignment. The attribute STOP, for instance, defines the class of all stop consonants and BILABIAL STOP would define the set/p, b, m/. Thus, a rule specifies a parameter value when a subclass of phonemes occur, in the context of another subclass. Such a formulation considerably reduces the number of rules. The classification as well as the rules are supplied as data to the system, giving it considerable flexibility. The spectrographic output of the concatenation stage is used to actuate a simulated series terminal analog synthesizer. Rudimentary prosodics are incorporated which modify a monotonous pitch contour with stress markers and interrogative or declarative termination of a sentence.