An Artificial Talker Driven from a Phonetic Input
John L. Kelly, Louis J. Gerstman · The Journal of the Acoustical Society of America · 1961
This paper describes a method of producing artificial speech from a phonetic input, i.e., symbols representing the names of phonemes corresponding to a given text are fed into a machine and the acoustic waveforms of connected speech emerge. The experimental work was accomplished on an electronic computer (IBM 7090), but the scheme is simple enough to permit realization with analog hardware. The talking machine program is divided into two parts. The first part simulates a more or less conventional resonance synthesizer of the tandem variety, requiring nine control signals; buzz intensity, hiss intensity, pitch, plus the center frequencies and bandwidths of three formants. Initially, this part of the program was used alone in experiments for which the inputs were detailed specifications of the control signals derived from spectrograms and physiological data, sampled at approximately three times the phonemic rate. Results from this phase were later combined with known results in speech perception to produce the rules used by the second program. That program accepts for its input the names of phonemes punched on IBM cards and produces the control signals to drive the resonant synthesizer.