A bachelor's thesis for ranking articulation disparity using 3D facial features generated by deep learning models

• Python
• Conda

• git clone this repository
• git submodule init and git submodule update to initialize the submodules
• Create conda environment voca & resolve dependencies from voca directory
• Create conda environment deca & install dependencies from deca directory
• Create conda environment autoeditor & run pip install auto-editor
• Create conda environment pyqt & run pip install pyqt5
• Activate pyqt environment and execute speech app

1. User provides input in the form of video
2. Frame rate of input video is changed to 24
3. Silence part is removed
4. Duration is adjusted
5. Extract Audio from Video
6. Convert Video to Frames
7. Convert Frames to 3D Meshes
8. Compare 3D Meshes with Standard

By default, this repository contains only one standard stream. If you wish to add more standard words, perform the following steps:

• Make sure selenium is installed
• Place your desired words in words.txt file to scrape from online dictionary.
• Run audio_scraper.py
• Place scraped mp3 files into standard audios folder
• Activate environment voca
• Run preprocess_audios.py
• Run generate_vertices.py
• You should see your standard's 3D Meshes generated by VOCA model