ObamaNet: Photo-realistic lip-sync from text

I don't care about this paper all that much. Getting the lip features from sound has interesting use for making lip sync available to indies.

• Standard Char2Wav to perform text to speech synthesis.
• LSTM network with time delay to predict mouth shape from audio features.
• Uses WORLD vocoder to get audio features from sound track.
• Uses dlib facial landmark detector to get key points of the face
• Uses principle component analysis to identify most important keys
• Creates a wireframe rendering of the lips and then uses image to image networks to synthesize the rest of the mouth from those images