Related Experiment Videos
Principal components representation of the two-dimensional coronal tongue surface
Eric Slud1, Maureen Stone, Paul J Smith
1Mathematics Department, University of Maryland, College Park, Md 20742, USA. evs@math.umd.edu
Phonetica
|September 17, 2002
Summary
Principal components analysis effectively represents English vowel tongue contours. This method simplifies complex articulatory data, revealing insights into vowel height and consonant influences.
Area of Science:
- Linguistics
- Speech Science
- Acoustic Phonetics
Background:
- Understanding tongue articulation is crucial for phonetics.
- Previous methods struggled to efficiently represent complex tongue movements.
- A need exists for dimensionality reduction in articulatory data.
Purpose of the Study:
- To apply principal component (PC) analysis for representing coronal tongue contours.
- To analyze English vowels in /s/ and /l/ consonant contexts.
- To assess the effectiveness of PCs in capturing articulatory variations.
Main Methods:
- Principal component analysis was performed on coronal tongue contours.
- Data from 6 subjects across 3 sessions with 5 replicates were used.
- Extensive preprocessing included curve alignment and normalization.
Main Results:
- Four principal components and a mean level accurately represented tongue curves.
- PC shapes were sensitive to data preprocessing (padding/truncation).
- PCs successfully reduced dimensionality and reflected vowel height and consonant context.
Conclusions:
- Principal component analysis is a viable method for articulatory phonetics.
- The approach can capture key physiological features of speech production.
- Further research should explore preprocessing's impact on PC interpretation.