Developing real-time low-power multi-channel speech separation algorithms

We are developing multi-channel speech separation algorithms for operation in the wild. The algorithms will be based on beamforming, deep learning solutions, and a multi-microphone ad-hoc wireless platform and robot developed within the group (see publications list at http://sensors.ini.uzh.ch/publications.html). The student will investigate ways of implementing multi-channel speech enhancement and source separation for reverberant environments.

Available Projects:
We offer MSc thesis projects and possible PhD position after the project. There is also a possible project with an industry group and with external academic experts. Projects that incorporate both audio and visual sensors including neuromorphic retina and cochleas are also available. Prerequisite for the project is prior knowledge in signal processing, machine learning and some deep network knowledge and/or knowledge on embedded hardware.