Parametrization of the vocal tract area function using a subset selection approach (L).

Area of Science:

• Acoustic phonetics
• Speech production modeling
• Computational linguistics

Background:

• Accurate vocal tract area functions are essential for understanding speech production.
• Previous models often require complex data or numerous parameters.
• A simplified, data-driven approach is needed for efficient vocal tract representation.

Purpose of the Study:

• To introduce a novel parametrization of the vocal tract area function.
• To identify a minimal set of independent points for accurate vocal tract shape reconstruction.
• To enable efficient and accurate modeling of vowel articulation.

Main Methods:

• Parametrization of the vocal tract area function using landmark point positions.
• Application of QR decomposition to area function data across various vowels.
• Identification of points with independent positional variations.

Main Results:

• Four key points (tongue body, lips, two on tongue back) were identified.
• These points capture the most independent positional patterns across vowels.
• Accurate reconstructions of vowel area functions were achieved using these four points.

Conclusions:

• A reduced set of points can effectively represent vocal tract geometry.
• This parametrization simplifies vocal tract modeling for speech research.
• The findings have implications for speech synthesis and articulatory phonetics.