Topic description Speech recognizers use a voice activation detector to remove the parts of the incoming signal that does not contain speech. The quality of this component is essential for robust speech recognition in adverse conditions such as in public places, cars, meeting rooms, ... The student will first study existing voice activation detectors and subsequently design and implement one and evaluate its performance. The criteria will be the false alarm rate (detect speech when there is none) and false rejection rate (declare speech as non-speech) under noisy conditions. The new method will be compared with an existing baseline. Experience with programming and some background in signal processing techniques are helpful skills. Contact person(s) Hugo Van hamme    (promotor) Kris Hermus