Inversion of articulatory-to-acoustic transformation in the vocal tract by a computer-sorting technique

Area of Science:

• Acoustic phonetics
• Speech science
• Computational linguistics

Background:

• Understanding the relationship between vocal tract articulation and acoustic output is crucial for speech production and perception.
• Previous models often simplify the complex, multidimensional relationship between articulatory parameters and acoustic features like formant frequencies.

Purpose of the Study:

• To develop and present numerical methods for inverting the articulatory-acoustic function, enabling prediction of vocal tract shapes from acoustic data.
• To investigate the phenomenon of compensatory articulation, where different vocal tract shapes produce similar acoustic outputs.

Main Methods:

• Developed a computational procedure for inverting the articulatory-acoustic function (y=f(x)) using computer sorting.
• Applied linearization techniques to explore local and larger regions of the articulatory-acoustic mapping.
• Synthesized sounds from various vocal tract shapes identified through the methods for perceptual evaluation.

Main Results:

• Successfully demonstrated a method to determine vocal tract parameters corresponding to specific formant frequencies.
• Identified 'articulatory regions' (fibers) that map to single points in the acoustic space, indicating acoustic-perceptual equivalence.
• Informal listening tests confirmed that many distinct vocal tract shapes can produce the same speech sound.

Conclusions:

• The developed numerical methods provide a powerful tool for analyzing the articulatory-acoustic relationship in speech.
• The findings highlight the significant potential for vocal tract shape variability in producing perceptually similar speech sounds.
• This has implications for understanding speech motor control, speech disorders, and developing more sophisticated speech synthesis and recognition systems.