Full Stack Lip Reader
Created Neural Network which could read lips. The model achieved over 97.5% validation accuracy, and included Convolutional layers, stacked LSTMs, Dense layers, Dropout, and Time Distributed Layers. Environment: Python, TensorFlow, Streamlit, Jupyter

Using just the frame by frame sequence of the person's lips (no audio), the Neural Network translates the lips to tokens, and from tokens to text.