Adaptive Speech Dereverberation based on Linear Prediction

Proposal for a Master Thesis


Adaptive Speech Dereverberation based on Linear Prediction


A speech signal recorded in a room will inevitably be affected by reverberant components produced by the reflections of sound waves on walls or objects in the acoustic environment. Reverberation is detrimental to the perceived quality of the observed speech signal and often makes human-machine interaction more challenging.

The goal of dereverberation is to reduce the energy of reverberant components in the recorded signal to lower the aforementioned detrimental effects. A powerful approach to dereverberation is the weighted prediction error (WPE) method, which requires no prior knowledge of the room transfer functions (TFs). If the desired sound source is moving, the algorithm has to adapt to the changes in the TFs, which can be accomplished by online algorithms.

The aim of this thesis is the evaluation of adaptive WPE algorithms starting with the implementation of [1]. This includes a literature survey and the investigation of promising approaches for different acoustical scenarios. As prerequisites, the student should have a good grasp of statistical signal processing and basic MATLAB programming experience.



Prof. Dr.-Ing. Walter Kellermann


M.Sc. Andreas Brendel, room 05.018,