Vocal-tract acoustics and speech synthesis

Basic researches in speech communications have significantly progressed in the past four decades. Paradoxically, the most of currently developing text-to-speech systems is based on a “cut-and-paste” waveform concatenation method. Its main advantage seems to come from the fact that detail knowledge about speech is not necessary. The “dumb” but efficient engineering approach is actually taking over a more scientifically sound rule-based synthesis in the development and application of a high quality text-to-speech system. We overview these two different approaches and then try to demonstrate the feasibility of a vocal-tract synthesis method which is nothing but a retake of an old speech synthesis method called “vocal-tract analog”. Some examples of vocal-tract synthesis of V₁CV₂ (vowel-consonant-vowel) utterances using an extremely simple phoneme concatenation by interpolation are demonstrated