Navigation

Coherence-based Multichannel Double-talk Detection

Proposal for a Master Thesis

Topic:

Coherence-based Multichannel Double-talk Detection

Description:

In a multichannel acoustic echo cancellation scenario, an active near-end speaker poses a serious challenge to the identification of the acoustic echo paths between loudspeaker and microphones, as the acoustic echo from the loudspeaker is mixed with the near-end
source in the microphone signal (‘double-talk’). This challenge is often resolved by halting the adaptation once near-end activity is detected. Many approaches have been proposed to detect double-talk in the single-channel case, e.g., as in [1]. Nowadays, most speech assistant devices, e.g., Amazon Echo, employ multiple microphones and loudspeakers, introducing the need for multichannel double-talk detection approaches. In this thesis, the approach in [1] shall be extended to the multichannel case, where the coherence between the multiple microphones in addition to the coherence between the far-end and the microphone signals is exploited to detect the occurrence of double-talk. Implementation and evaluation should be done in MATLAB.

Download-DTD

Professor:

Prof. Dr.-Ing. Walter Kellermann

Supervisior:

M.Sc. Mhd Modar Halimeh (Cauerstr. 7, room 5.13, mhd.m.halimeh@fau.de)

Prerequisites:

Course ‘Digital Signal Processing’, MATLAB experience.

Available:

Immediately