Motivation:

Robots are becoming more and more integrated into our everyday lives. In this sense, it is important to make human-machine interaction as natural as possible.

Visual Voice Activity Detection (VVAD) is an important cognitive function that allows people to see if the interaction partner is speaking. Robotic systems repeatedly show deficits in this aspect, which means that interaction with the robotic system is unnatural and not very fluid. In the thesis, a VVAD system is to be developed on the basis of the VVAD-LRS3 data set. 

Goal:

≫    Development of VVAD system for a robotic system

Prior Knowledge:

≫    Python

≫    Deep-Learning

Related Work:

≫    Adrian Lubitz, Matias Valdenegro-Toro, Frank Kirchner (2021) 
The VVAD-LRS3 Dataset for Visual Voice Activity Detection,                    https://doi.org/10.48550/arXiv.2109.13789

We look forward to receiving your complete informative application documents including current transcript of grades. For further information and application, please contact:  Adrian.Lubitz@dfki.de.