A modular architecture for articulatory synthesis from gestural specification.

Area of Science:

• Speech Synthesis
• Articulatory Phonetics
• Biomedical Engineering

Background:

• Articulatory speech synthesis requires integrated models of the vocal tract, glottis, and aerodynamics.
• Previous models often lack modularity and integration of real-time imaging data.

Purpose of the Study:

• To propose a novel modular architecture for articulatory speech synthesis.
• To integrate diverse models for vocal tract, glottis, aero-acoustics, and control.
• To validate the architecture using speaker-specific data and synthesis tasks.

Main Methods:

• Developed a modular architecture combining statistical articulatory models (from real-time MRI data) with aerodynamic and glottal models (Maeda's work).
• Implemented an articulatory control module using dynamical systems inspired by task dynamics.
• Utilized an alpha-beta model for midsagittal to area function conversion.

Main Results:

• Successfully synthesized vowel-consonant-vowel sequences with plosive consonants.
• Demonstrated speaker-specific modeling by building and simulating models for two individuals.
• Validated the modular approach for articulatory speech synthesis.

Conclusions:

• The proposed modular architecture provides a flexible and effective framework for articulatory speech synthesis.
• Integration of statistical articulatory models with aerodynamic and control systems enhances synthesis realism.
• The approach is adaptable for simulating different speakers and their articulatory behaviors.