Synthesis of Speech From a Dynamic Model of Vocal Cords and Vocal Tract

Speech sounds can be synthesized by a variety of means used to construct signal waveforms. Many ingenious methods have been recorded. But speech synthesis generally has the practical purpose of producing intelligible sounds from control data that are as parsimonious as possible. In other words, the control data should represent an efficient, concise coding of the speech information. This motivation applies as much to analysis/synthesis techniques for speech transmission as to computer voice-response systems which strive for efficient vocabulary storage and high versatility in message fabrication. Because speech is a human-generated signal, it is unlikely that a synthesis method can achieve the ultimate parsimony of input control without considerable attention to the parameters a human overtly manipulates in speaking. That is, one increases the information "built into" the synthesizer when its design exploits fundamental properties of the human speech mechanism. We therefore have chosen an approach to synthesis with which we can identify overtly the significant physiological parameters important 485