Dynamic acoustic vowel distances within and across dialects.

Area of Science:

• Phonetics and Acoustics
• Sociolinguistics
• Computational Linguistics

Background:

• Vowel acoustic similarity varies significantly across regional American English dialects.
• Traditional acoustic measures often use discrete time points, potentially missing dynamic formant trajectory information.
• Understanding dialectal differences in vowel perception is crucial for linguistic research.

Purpose of the Study:

• To apply and compare novel language and accent distance measures for evaluating vowel category similarity within talkers.
• To examine the utility of these measures in capturing predicted regional dialectal variations in American English.
• To assess the effectiveness of dynamic time warping and formant trajectory analysis in dialectology.

Main Methods:

• Dynamic time warping (DTW) of mel-frequency cepstral coefficients (MFCCs) to assess spectral acoustic distance over time.
• Root-mean-square (RMS) distance and generalized additive mixed models (GAMMs) applied to selected formant trajectories.
• Comparison of acoustic distances across Southern, New England, and Northern American English dialects.

Main Results:

• DTW of MFCCs successfully captured predicted Southern American English vowel similarity.
• RMS distance and GAMMs captured predicted vowel similarity patterns for Southern, New England, and Northern dialects.
• GAMMs explained the most variation but do not provide a single distance value, unlike DTW and RMS.

Conclusions:

• Dynamic acoustic measures, including DTW and formant trajectory analysis, are valuable for studying dialectal vowel variation.
• Generalized additive mixed models offer a nuanced approach to capturing complex dialectal patterns.
• These methods enhance our understanding of acoustic vowel category similarity across diverse American English dialects.