Vowel and consonant contributions to vocal tract shape

Area of Science:

• Speech Science
• Articulatory Phonetics
• Computational Linguistics

Background:

• Understanding the complex dynamics of speech production is crucial for developing advanced speech synthesis and recognition technologies.
• Existing models often struggle to accurately capture the interplay between vowels and consonants in articulatory terms.
• X-ray microbeam data provides precise articulatory information but requires sophisticated methods for interpretation.

Purpose of the Study:

• To develop a computational method for decomposing vowel-consonant-vowel (VCV) utterances into articulatory components.
• To model vowel sequences as time-dependent perturbations of the neutral vocal tract shape.
• To represent consonants as superposition functions that modify vocal tract shape over time.

Main Methods:

• Utilized x-ray microbeam articulatory data from VCV utterances ([pa], [ta], [ka]).
• Developed a model separating VCVs into vowel-to-vowel transitions and consonant superposition functions.
• Employed canonical deformation patterns to represent vocal tract shape changes.
• Quantitatively assessed model accuracy using root-mean-square (rms) error and correlation coefficients.

Main Results:

• The developed model successfully separated VCV utterances into distinct articulatory components.
• Reconstructed VCV utterances ([pa], [ta], [ka]) showed reasonable approximations of the original sounds.
• Quantitative analysis confirmed the model's accuracy through low rms error and high correlation coefficients.

Conclusions:

• The proposed method provides a robust framework for analyzing and modeling speech articulation.
• This approach enables a more detailed understanding of consonant-vowel interactions in the vocal tract.
• The established method lays the groundwork for future research on diverse speech materials and speaker variations.