Skip to contents

Vowel space analysis (F1×F2 area ratio)

lst_vowel_space.Rd

Computes vowel space area from formant data using k-means clustering and reference vowel comparison. The vowel space ratio is computed as the ratio of the actual vowel space convex hull area to a reference vowel space area for the same gender.

Usage

lst_vowel_space(formant_data, gender = 1, mode = "triangle", scaling = FALSE, plot_formants = FALSE, return_jstf = FALSE)

Arguments

formant_data

Data frame or matrix with columns F1, F2 (or first two columns). F1 and F2 should be in Hz.

gender

Speaker gender for reference vowel selection: 0 = female, 1 = male (default), 2 = child

mode

Vowel space computation mode: "triangle" = use 3 corner vowels (default, more robust) "polygon" = use 4 vowels (may be less stable)

scaling

Apply Bark frequency scaling (default: FALSE)

plot_formants

Show scatter plot of formant points (default: FALSE)

return_jstf

Logical. Return JsonTrackObj instead of data.frame? Default FALSE. When both toFile and return_jstf are TRUE, the file is written AND the object returned.

Value

List with elements:

  • vowel_space_ratio: Numeric, ratio of actual to reference vowel space area (0–2+ range)

  • centroids: Matrix of estimated vowel formant centroids from k-means clustering

  • n_frames: Number of frames used in computation

Details

Algorithm:

  1. Filter formant data by gender-specific frequency ranges to identify candidate vowels

  2. Run k-means clustering using reference vowel centroids as initialization

  3. Match estimated centroids to reference vowels using Mahalanobis distance

  4. Compute convex hull of matched vowel points (actual vowel space)

  5. Compute convex hull of reference vowels

  6. Ratio = actual area / reference area

Interpretation:

  • Ratio = 1.0: perfect agreement with reference vowel space

  • Ratio < 1.0: speaker has reduced vowel space (dysarthria, age, etc.)

  • Ratio > 1.0: speaker has expanded vowel space (hyperarticulation, singing)

  • NaN/error: insufficient vowel tokens (<1000 frames) in valid range

Reference vowel sets:

  • Triangle: /i/, /a/, /u/ (corner vowels)

  • Polygon: /i/, /ε/, /a/, /u/ (4-vowel system)

Minimum frame requirement: 1000 frames must be within gender-specific frequency bounds, or function returns 0 ratio (insufficient data).

Examples

if (FALSE) { # \dontrun{
# Compute vowel space from file batch
files <- c("speaker1.wav", "speaker2.wav")
formants <- trk_formant_burg(files, toFile = FALSE)

# Extract F1-F2 and compute vowel space
formant_df <- data.frame(F1 = formants$fm1, F2 = formants$fm2)
vs <- lst_vowel_space(formant_df, gender = 1)
cat("Vowel space ratio:", vs$vowel_space_ratio, "\n")

# For female speaker
vs_female <- lst_vowel_space(formant_df, gender = 0)

# Using Bark scaling
vs_bark <- lst_vowel_space(formant_df, gender = 1, scaling = TRUE)
} # }