Piecewise-planar vowel formant distributions across speakers

Area of Science:

• Acoustics
• Linguistics
• Speech Science

Background:

• Previous research identified a two-part piecewise-planar surface for a single speaker's vowel formant frequencies.
• This study investigates the generalizability of this finding across multiple speakers.

Purpose of the Study:

• To extend the analysis of vowel formant frequencies to five additional speakers.
• To determine if a similar two-plane representation exists for these speakers.
• To assess the degree of similarity and variation in these representations across speakers.

Main Methods:

• Analysis of the first three vowel formant frequencies (F1, F2, F3) for six speakers.
• Fitting a two-part piecewise-planar surface to the formant data for each speaker.
• Statistical analysis of the rms spread and plane orientations compared to average planes.

Main Results:

• A similar two-plane representation was found for all five additional speakers, with rms spreads between 69 and 103 Hz.
• Significant departures from average plane orientations were observed for individual speakers (average angles of 10° for front vowels, 13° for back vowels).
• Speakers grouped into two subsets exhibiting near-uniform scalability, but strict three-dimensional uniform scaling was rejected.
• Third formant (F3) values for retroflex vowels were generally lower than predicted by the planar models.

Conclusions:

• While individual vowel formant frequency patterns show significant variation, they adhere to a general piecewise-planar structure.
• The hypothesis of strict uniform scaling across all speakers in three dimensions is rejected.
• The findings suggest a complex interplay of individual vocal tract characteristics and acoustic principles in vowel production.