In polvanrijn/contouR: Compute dynamic features on pitch contours

Set the right variables and load the package

base_dir = "/Users/pol/owncloud/HiWi_folder/02_running_projects/02_MA_thesis/01_Data/Preprocessed/"
Fujisaki_path = paste0(base_dir, "Fujisaki/")
pt_path = paste0(base_dir, "PitchTiers/08_manual_corrected/")
library(contouR)

The first step is to convert the PitchTiers into f0_ascii files.

files = list.files(pt_path)
files = files[grepl(".PitchTier", files)]
if (!dir.exists(Fujisaki_path)){
  dir.create(Fujisaki_path)
}

for (pt_name in files){
  pt = read_PitchTier(paste0(pt_path, pt_name))
  filename = strsplit(pt_name, "\\.")[[1]][1]
  write_f0_ascii(pt$t, pt$f, paste0(Fujisaki_path, filename, ".f0_ascii"))
}

Now convert the f0_ascii to the PAC files with interpolate.exe by Mixdorff. What I recommend to do is to use a Windows VM (I used Windows 10) and setup SSH.

You can run this by running the following command on Windows

@echo off & setlocal

cd Z:\01_Data\Preprocessed\Fujisaki\
for /r %%i in (*.f0_ascii) do (
    cd Z:\02_Scripts\Bash
    for /f "tokens=1 delims=." %%a in ("%%i") do (
      interpolation %%a%.f0_ascii 0 4 1e-006 auto 2
    )
)

In the next step, we'll look at the optimal alpha and beta values. This is done with explore_ft_space.

top_scores = explore_ft_space(Fujisaki_path, pt_path, plot = FALSE)
write.csv(top_scores, "top_scores.csv", row.names = FALSE)

Phrase command

From these optimal hyper parameter combinations we can compute features on the phrase command

top_scores = read.csv("top_scores.csv")
phrase_features = compute_phrase_features(top_scores, Fujisaki_path)
head(phrase_features)

We see that both slope computations (and also the RMSE) are highly correlated. Also we find a negative correlation between the amplitude and the slope

library(corrplot)
corrplot(cor(na.omit(phrase_features[, 3:8])), method="circle", type = "upper")

Visualization

plot_optimal_estimation("SL_DIS_VX1b", pt_path, Fujisaki_path)

Accent command

With the following command we can check the accent command

accent_features = compute_accent_features(top_scores, compute_accent_features)
head(accent_features)

The correlation plot reveals that the amplitude of the first accent component is positively correlated with the last compontent.

corrplot(cor(na.omit(accent_features[, 2:5])), method="circle", type = "upper")

polvanrijn/contouR documentation built on Feb. 10, 2020, 3:45 a.m.