An articulatory-formant speech synthesizer via a neural network

摘要

An articulatory-formant synthesizer oriented toward implementationby a neural network is proposed. Speech output is the sum of the formantfilters' responses terminated by an appropriate source of excitation. Amodel of articulatory dynamics is adopted for the synthesizer. A fastprocedure for eigenfrequency calculation which makes it possible toreduce the computational cost in the synthesis to about three millionoperations per second of speech is presented. The synthesizer wassimulated on a PC in FORTRAN. The model of the syntheses appears toaccommodate a neural network implementation. A procedure for trainingthe neural network is proposed