1 of 1

Multidimensional Vowel Dispersion as a Cue to Stylistic Variation

Julian Vargo1 & Akul Shivkumar

1Department of Spanish & Portuguese, University of California, Berkeley

  • Force Aligned 30 minutes of speech by Tim Cook using the Montreal Forced Aligner (McAuliffe et al 2017)
  • Several recordings of Cook exist in various lengths, with various speech styles.
  • Vowel dispersion can extend beyond F1 and F2.

​

  • Vowel dispersion is constantly shifting, so plotting dispersion change throughout the interview can be insightful.

​

  • Moving away from vowel

space area can provide more

nuance in sociolinguistic

research.

Conclusion

METHODOLOGY

Results

INTRODUCTION

  • Vowel space area (VSA) is associated with stylistic or accommodative speech shifts.
  • Larger areas signify greater hyperarticulation (Ferguson & Kewley-Port 2007).
  • VSA is typically calculated as the area between peripheral phoneme midpoints such as /i/, /u/, and /a/ (Lindblom 1990; Burdin, Turnbull & Clopper 2014).
  • Story & Bunton (2017) furthered vowel area methods through density-controlled area calculations with convex hulls.
  • The casual interview yielded the smallest vowel-space volume (0.1589 ln[Hz]3)
  • The MIT commencement speech had the highest volume (0.2579).
  • Dispersion GAM reveals that overall dispersion increases at moments where high-stakes messages occur, such as selling a new product.
  • Drops in dispersion signal a joke or casual situation.
  • GAMs are only meaningful after phoneme-centered normalization, signaling a need for improved dispersion techniques.

​

1) We develop a 3D density-controlled vowel volumetric analysis for F3 variation.

​

2) We develop a method to analyze vowel dispersion fluctuations for discourse analytical contexts.

julianvargo@berkeley.edu

Citations

  • Low-density vowel removal changes as audio length increases.
  • Density cutoffs must be set at a certain density percentile, to control for time.
  • Vowel space volume is a single metric taken over time.

3D Vowel Space Volume

Our study improves current vowel area limitations:

Midpoint-based area:

reductionist

Convex hull area:

improved accuracy

The same (hypothetical) vowel space, with two different area measurements.

2) Dispersion Contours

  • Each vowel was grouped by phoneme and normalized.
  • Each vowel assigned a distance from center of the vowel space.
  • GAM trajectories measure dispersion change throughout an audio.

1) 3D Vowel Space

  • Vowels were placed on an F1 by F2 by F3 scatterplot.
  • The local density of a vowel in 3D space was calculated.
    • If a vowel fell below a certain density, then it was removed.
  • Convex hull wrapped around the solid, and the volume was calculated.

i

u

æ

ɑ

ɪ

ʊ

ɛ̠

ɔ̟

Large Vowel Area

=

Tense,

formal,

careful speech

Small Vowel Area

=

Lax,

casual,

spont. speech

Dispersion is measured for each phoneme independently. Lighter color = more dispersion

Vowel Dispersion GAM Trajectories

  • Dispersion GAMs can be used for measuring detailed stylistic shifts.
  • Key moments of each audio show large changes in the graph

Template ID: debatingdenim Size: 48x36