-
Notifications
You must be signed in to change notification settings - Fork 2
a toolkit for prosody processing using HTK, Praat, and WEKA
spirineta/protk
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
This is ProTK v0.2.2 ==================== Release Information ------------------- Author: T. Christie and Serguei Pakhomov Contact: tchristie@umn.edu Organization: University of Minnesota Last Modified: August 9, 2011 License Information ------------------- This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA What is ProTK for? ------------------- See Usage, and the included paper. Installation ------------------- ProTK v0.2 is written to work in a Mac/Linux environment. 1) REQUIRED SOFTWARE: PRAAT First, you should install Praat: http://www.fon.hum.uva.nl/praat/ or with Linux: apt-get install praat The expected install path is: "/Applications/Praat.app/Contents/MacOS/Praat" If you are running Linux or have Praat installed in a different location, you can change the path with a command-line flag (see below) IMPORTANT: The Praat script included in ProTK works with Praat 5.1, 5.1.78, and 5.2.13. The latest version of Praat for OS X (5.2.32) appears to have changed the way scripts are parsed, so the included script does not work. PYTHON ProTK is written to work with Python v2.6. Plotting will not work with version 2.7, but everything else should. You can downlost Python v2.6 here: http://www.python.org/download/releases/2.6.6/ NUMPY NumPy is also used for numerical computations. It can be downloaded here: http://sourceforge.net/projects/numpy/ Be sure to download a version of NumPy compatible with Python 2.6. As of this writing, v2.6 is the latest version of Python that NumPy supports. You can check that everything is installed correctly by (on OSX/Linux) > python2.6 >> import numpy If you get no errors, you're good to go. You can close Python by typing quit(). MATPLOTLIB Matplotlib is required for plotting. You can run ProTK without the plotting option by leaving out the -g flag when running. You can download Matplotlib here: http://sourceforge.net/projects/matplotlib/files/matplotlib/matplotlib-1.0/matplotlib-1.0.0-python.org-py2.6-macosx10.3.dmg/download for OSX. 2) PACKAGE Several folders were included in the zip file: ProTK_v0.2.2_code/ This contains the Python code for the program wav/ This is where you should put your .wav files to be processed RENAME_*/ This is where your .txtgrid files go. ProTK is designed to create training and testing sets for WEKA, a machine-learning classifier. Therefore, the "true", reference textgrid files should go in one folder and the files to be evaluated go in the other. ProTK expects each .txtgrid file to have "phone" and "word" tiers, even if they both contain phoneme information. An example file is included in the sample_files folder, called EXAMPLE.txtgrid. You should change the "RENAME" folder names to reflect the type of recognition the processes have undergone. Usage ------------------- To run ProTK you need at least one set of WAV and corresponding Txtgrid files in subfolders of the ProTK directory (the WAV files must be in wav/ but you can call the textgrid file whatever you want - say "truth). These Txtgrid files should have two tiers: a phone tier and a word tier. In the "truth" condition, the words in the word tier should be labeled as FILLEDPAUSE_UM or FILLEDPAUSE_AH in the case of filled pauses. These will be used to mark filled pauses in the truth AND any other source folder. In the terminal, change directories to the directory containing the code (ProTK_v0.2.2_code - changing directories isn't strictly necessary). From there, you can type: python Tk.py -h or python2.6 Tk.py -h (if your default Python is not 2.6) to see the command line options. The available options and their default values are: directory = "../", configLocation="../", numProcesses = 1, truth = "manual", beforeAfter = 1, writeToFile = False, cleanOldResults = False, extractPraat = False, reformat = False, graph = False writeToFile = False, quiet = False, source = False, praatLocation = "/Applications/Praat.app/Contents/MacOS/Praat", wekaOnly = False, wekaname = "", tags = "FILLEDPAUSE", tagtier = "word", prosodytier = "phone" Options: -h, --help show this help message and exit -d DIRECTORY, --directory=DIRECTORY REQUIRED: full path to the directory to be processed --tags=TAGS optional: names of things to tag. Default is filledPauses. This is a placeholder - functionality for this will be added in a future release. -t TRUTH, --truth=TRUTH optional: where the 'truth' txtgrid files are. Default is 'manual'. -s SOURCE, --source=SOURCE optional: subfolders to process & make a WEKA file for. Type case-sensitive foldernames separated by commas with NO spaces. The default is auto. -p NUMPROCESSES, --processes=NUMPROCESSES optional: The number of processes to run simultaneously during feature extraction. The default is 1. -c CONFIGLOCATION, --configLocation=CONFIGLOCATION optional: path to the config file, if it's not in the directory of the project -b BEFOREAFTER, --beforeAfter=BEFOREAFTER optional: integer specifying how many pre/post intervals to include in WEKA file. Default is 1 (meaning 1 before, 1 after) -e, --erase optional: Erases previous extraction results by deleting and recreating db, arff, extracted, praatOutput and formatted directories -x, --extractpraat optional: extract Prosody information using Praat. Default is to only do this if being run for the first time. NOTE: -x does not require an additional argument. -r, --reformat optional: force reformatting of info extracted from Praat. The default is to to do this only if being run for the first time. Doesn't require an additional argument. -g GRAPH, --graph=GRAPH optional: Plots intervals and context. Include a positive number to include context surrounding the graph. ***IMPORTANT***:if you use this option, -p will be set to 1 due to a limitation in MatPlotLib. -o, --writetofile optional: breaks up extracted info and writes it to text files. Default is to NOT do this. -q, --quiet optional: if used, suppresses all screen output except error messages. -w WEKANAME, --wekaname=WEKANAME optional: the name of the weka file to be output. -z, --wekaonly optional: ONLY do weka files, no extraction --sphinx optional: MUST USE if files to be processed include output from Sphinx. Tells the reformatter to expect milisecond time stamps in the phoneme file. --praatpath=PRAATLOCATION optional: path to praat executable if not in default OS X path. In Linux, it may be /usr/bin/praat --tag=TAGS optional: specify string to search for indicating a particular tag. Default is FILLEDPAUSE --tagtier=TAGTIER optional: specify which tier contains tagged datastring to search for indicating a particular tag. Default is FILLEDPAUSE --prosodytier=PROSODYTIER optional: specify the relevant tier for prosodic analysis. The default is 'phone'. Suggested uses: A typical usage is: python Tk.py -t RENAME_truth -s RENAME_auto -p 4 -g 1 -e -x -o This erases old results (-e), extracts Praat info (-x), reformats and writes prosodic info to file (-o), plots results (-g 1). Note that although (-p 4) is selected, graphing requires that p=1, so the value is automatically changed to 1. Using an existing project (where info has already been extracted and reformatted), to re-calculate the WEKA ARFF files in the "new" directories, type: python Tk.py -s new,new2,new7 -t MANUAL -z *******IMPORTANT******** - if processing *.wav files for the first time, you must use the -x flag to do initial Praat extraction. - if plotting, limitations in the Matplotlib package require that only one process is run at a time. Therefore, p is set to 1 automatically.
About
a toolkit for prosody processing using HTK, Praat, and WEKA
Resources
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published