Navigation

Double-Talk Detection for Robot Audition

Proposal for a Master Thesis

Topic:

Double-Talk Detection for Robot Audition

Description:

In an Acoustic Echo Cancellation (AEC) problem, the performance of an acoustic echo canceler is usually impaired by the presence of a near-end interferer. This double-talk situation is often addressed using Double-Talk Detectors (DTD), which stop the adaptation of the acoustic echo canceler as soon as double-talk is detected. However, in time-varying environments, DTDs should also be able to distinguish a change in the echo path from the presence of a near-end signal.
In a robot audition scenario, the near-end interferers are not limited to other speakers only, but also include the robot self-noise. In addition, due to the robot movement, the acoustic path between the microphone and the loudspeakers is also time-varying.

In this thesis, the use of coherence-based, and artificial neural networks-based DTDs for robot audition shall be investigated and evaluated. Implementations are expected to be done in MATLAB.

Download

Professor:

Prof. Dr.-Ing. Walter Kellermann

Supervisior:

MHD Modar Halimeh, M.Sc.,
Alexander Schmidt, M.Sc.,
(Cauerstr. 7, room 5.13, mhd.m.halimeh@fau.de, alexander.as.schmidt@fau.de)

Prerequisites:

Course ‘Digital Signal Processing’, MATLAB experience.

Available:

Immediately