Development of a deep learning based lipreading system for patients with speech impairment

data science and Machine Learning Diplomarbeiten (never - cancelled)


Arne Peine, Anke Schmeink,


In recent years, there has been a quantum leap in the performance of visual speech recognition systems, thanks to the advances in deep learning techniques. In this Master’s thesis, you will evaluate different approaches (e.g. LSTMs; fully convolutional models; a recently proposed transformer model) towards their performance on predicting speech from visual information. You will develop a system for gathering training data and apply your models on real patient data, potentially helping patients with speech impairment.

Download information as pdf.