Dynamic 3-D visualization of vocal tract shaping during speech

Area of Science:

• Speech Production and Acoustics
• Medical Imaging and Visualization
• Computational Linguistics

Background:

• Noninvasive imaging is crucial for speech research, but current methods lack 3D dynamic visualization of the vocal tract.
• 3D dynamic Magnetic Resonance Imaging (MRI) offers potential for comprehensive vocal tract analysis.
• Understanding vocal tract dynamics is key to advancing speech production research.

Purpose of the Study:

• To develop a novel method for creating 3D dynamic movies of vocal tract shaping.
• To enable detailed visualization of articulatory movements during speech production.
• To improve the temporal resolution of vocal tract imaging.

Main Methods:

• Acquisition of multiple sagittal 2D real-time MRI movies with synchronized audio recordings.
• Temporal alignment of audio data using mel-frequency cepstral coefficients (MFCC).
• Alignment of 2D MRI sequences using dynamic time warping (DTW) for synthesized 3D visualization.

Main Results:

• Successfully generated 3D dynamic movies of vocal tract shaping from 2D MRI data.
• Visualized salient aspects of lingual articulation, including tongue grooves and sublingual cavities.
• Achieved a temporal resolution of 78 ms for dynamic 3D vocal tract visualization.

Conclusions:

• The novel method effectively creates 3D dynamic visualizations of the vocal tract during speech.
• This technique advances the study of speech production by providing detailed articulatory insights.
• The method holds promise for future research in speech science and related fields.