Python toolkit for articulatory phonetic analysis of image data: lingual ultrasound and video of facial landmarks. Facial landmark detection is carried out using dlib
on images preprocessed with opencv3
and ndimage
. (Ultrasound feature extraction TBA.) Dimensionality reduction is carried out using scikit-learn
.
Files in the scripts
directory also carry out forced alignment and acoustic analysis routines from various stages of writing my dissertation; these and the ultrasound analysis routines they coincide with will eventually be converted to functions in a sensible namespace.
The following dependencies, among others, are required for facial feature extraction using the functions and classes in imgphon/imgphon
: opencv3
, imutils
, and dlib
. The dependencies for dlib
are especially large and complex. We have used the installation instructions for dlib on Unix-type systems at bit.ly/2HEhCQP with some success. The instructions provided at davisking/dlib
may also help.
To configure upon downloading the package, simply navigate to the root directory of the repo and run python setup.py install
.
A demonstration of lip shape extraction and dimensionality reduction can be found in lip-demo.ipynb. A demonstration of some tools for extracting and processing ultrasound video frames from the Telemed EchoBlaster's binary outputs can be found in echob-data-demo.ipynb. Use of these extracted features for analysis coming soon.
Coming soon