Colloquium - Details

Master-Vortrag: Estimation of the Geometry of Microphone Arrays on Mobile Devices

Ahmad Hamad
Montag, 16. Januar 2017
11:00 Uhr
Hörsaal 4G

Microphone array systems are involved in a wide range of applications like speech enhancement and source localization and identification. The ability of a microphone array to perform spatial filtering (beamforming) can be severely degraded by uncertainty concerning array shape. Thus, estimating the array shape is a critical task. The goal of the work of this thesis is to blindly estimate the shape of the microphone array. The term "blind" means that the estimation process relies only on the signals that are available at the input of each microphone.  

In this thesis, two approaches to array shape estimation are investigated: The Diffuse Noise Coherence (DNC) approach and the Time Difference of Arrival (TDOA) approach. The diffuse noise coherence can be exploited to directly estimate the pairwise distances between pairs of microphones. Such distances can then be used to estimate the array shape by a process called Classical Multi-Dimensional Scaling (CMDS). However, the assumption of diffuse noise is rather an overly simplified one. In realistic room environments, acoustic events’ signals rather than diffuse noise can be exploited by firstly estimating the time delays of between pairs of microphones. Such time delays can be used to estimate the array shape. Two approaches to exploit the time delays to estimate the array shape are discussed: The distance cube approach and the orthogonal projection approach. The distance cube approach requires multiple sources to estimate the array shape while the orthogonal approach, which is a novel contribution of this thesis, requires two sources to estimate the shape of planar arrays and three sources for the three-dimensional arrays. This thesis also proposed a method to detect and correct outliers in the estimated pairwise distances.

back