Scripts to accompagny 
"Automatic tracheoesophageal voice typing using acoustic parameters"

Use:
Linux   - bash
OSX     - X terminal, e.g., XQuartz (http://xquartz.macosforge.org/landing/)
Windows - Cygwin/X (http://x.cygwin.com/)

Download:
TEVA - http://www.fon.hum.uva.nl/IFA-SpokenLanguageCorpora/NKIcorpora/NKI_TEVA/
Praat - http://www.praat.org
Praat and TEVA have extensive help available in the program and on their 
web sites, e.g., at TEVA->Settings->Manual.

R is available from http://cran.r-project.org/
There is extensive documentation available for R

Workflow (based on command line interpreters)

- Start with a directory containing the recordings you want to evaluate

Open the directory in TEVA->Settings->Directory
Save the list in TEVA->Settings->Save

Process all recordings in TEVA, selecting a window and entering one of
the AST classes (1-4). When you press one of the buttons 1-4, the AST 
class and the window are saved. If you press 9, the window is saved but 
no class is stored. 0 will wipe this information for this file. The list 
is automatically saved. "example.Table" is an example of such a table.

After completing the AST classification, or even a consensus 
classification, open AcousticMeasureScripts.praat in Praat and Run->Run.
Enter the name of the table and the output file in the Form. Press OK.

Enter the name of the resulting output file in LOOC_data.R or 
Bootstrap_data.R. Run execute_all_classification_analysis.sh or 
execute_all_regression_analysis.sh with LOOC_data.R or Bootstrap_data.R

The output contains a lot of "messages" that can be filtered by adding
"2>/dev/null" to the command line.

Contains:

- AcousticMeasureScripts.praat
  Praat script that takes a file or files and outputs a table with 
  acoustic measurements as described in the paper.
  More information can be found in the script.
  Use:
  Open in Praat for interactive use or run from the command line:
  praat AcousticMeasureScripts.praat <in-filename> No \
           <StartTime> <EndTime> <out-filename> 0
  
- model_AST.R
  R script that will model AST classification as a function of the
  acoustics. It can use one of 7 machine learning algorithms
  lm, lda, qda, svm, randomforest, rpart, nnet
  AST = F(acoustics)
  More information can be found in the script.
  Libraries: 
  gplots, irr, rpart, CORElearn, randomForest, e1071, ndl, nnet
  Use:
  R --vanilla --slave -f model_AST.R --args lm '(Linear model)' classification [args]
  
- LOOCV_data.R
  Settings file to run modelAST.R using LOOCV and different training/test
  data
  
- Bootstrap_data.R
  Settings file to run modelAST.R using Bootstrap and different 
  training/test data

- execute_all_classification_analysis.sh
- execute_all_regression_analysis.sh
  Run model_AST.R for all algorithms in classification or regression mode.
  Use:
  bash execute_all_classification_analysis.sh LOOCV_data.R

- example.Table
  An example tsv table as obtained from TEVA
  Uses examples from:
  http://www.provoxweb.info/sounds.html
  
- example_acoustics.tsv
  An example tsv table as obtained from applying 
  AcousticMeasureScripts.praat on example.Table
  
- Speech
  Directory with 4 example recordings from 
  http://www.provoxweb.info/sounds.html
  These examples have been used to generate example.Table
  and example_acoustics.tsv