Studentische Arbeit - Details

For multiple applications of headphones and earbuds, an estimate of the user's own voice is of great importance. These include improved quality of transmitted speech during phone calls, better speech recognition, but also reduction of the occlusion effect by playing back the user's own voice where it is attenuated by the headphone. As modern headphones and earbuds are often equipped with multiple microphones per side, the question arises, how we can combine the resulting microphone signals for the goal of own-voice estimation.

Different algorithms have been described in the literature to utilize multiple microphones for speech enhancement. These include beamforming and multi-channel noise reduction among others. The main target of this thesis is to characterize the multi-microphone system of the headphone and investigate, if we can adapt selected speech enhancement techniques for this system.

First, the literature shall be reviewed for existing techniques. Then, the multi microphone system of an exemplary headphone shall be characterized by measurements with a dummy head. Based on the measured transfer functions, selected techniques can be evaluated via simulation. Special attention should be payed to the robustness of these methods under uncertainty. Then, the existing techniques shall be extended if possible, e.g. to realize an advanced transparent own voice hearthrough functionality.