Tidy Vowel Normalization

Setup and plot original data

library(tidyverse)
library(khroma)
theme_set(theme_minimal())

library(untidydata)

tidy_vowels <-
  spanish_vowels |>
  separate(
    col = label,
    sep = "-",
    into = c("id", "gender", "vowel")
  ) |> 
  # Recoding gender labels
  mutate(
    gender = case_when(
      gender == "female" ~ "women",
      gender == "male" ~ "men"
    )
  )

tidy_vowels |> 
  ggplot(aes(f2, f1, color = gender)) +
    stat_density2d()+
    scale_x_continuous(trans = "reverse")+
    scale_y_continuous(trans = "reverse")+
    scale_color_brewer(palette = "Dark2")

Unnormalized vowels

tidy_vowels |> 
  ggplot(aes(x = gender))+
    geom_violin(aes(y = f1, fill = "f1"))+
    geom_violin(aes(y = f2, fill = "f2")) +
    scale_y_continuous(trans = "log10")+
    scale_fill_bright()+
    labs(y = "frequency (hz)")

Normalizing

z-scoring a.k.a. “Lobanov”

Log-mean

Log-mean normalization, following Barreda (2021).

References

Barreda, Santiago. 2021. “Perceptual Validation of Vowel Normalization Methods for Variationist Research.” Language Variation and Change 33 (1): 27–53. https://doi.org/10.1017/S0954394521000016.

Reuse