Acoustic Source Localization is of interest in a variety of applications, ranging from teleconferencing and hands-free human computer interaction to pro-audio. In this project, an application-oriented approach to planar Acoustic Source Localization using a high quality coincident microphone array is researched. Multiple processing blocks are implemented in order to generate a reactive, yet stable Direction of Arrival estimation tuned toward speaker tracking. Building on an energy based scanning method, individual characteristics, such as sound field directivity and static sound source positions are used for adaptive smoothing of the detected angle. In this paper
, the methods and resulting performance gain are discussed for the individual components of the algorithm. This publication
focuses on the use of machine learning components to increase ASL stability in noisy environments and domain-specific target signal characteristics. Audio quality is assessed using listening tests, which are presented here
and show a significant increase in subjective sound quality, noise suppression, and speech intelligibility when combining the tracker with a beamforming algorithm for coincident microphone arrays.
The following demo contains recordings of a virtual conference scenario, provided at three levels of processing. The first audio example contains the omnidirectional signal, synthesized using front-facing and rear-facing cardioid microphones. The second signal contains a 1st order supercardioid beam, created using the localizer algorithm and gradient synthesis. The last signal contains a high-directivity beam created using the adaptive filtering algorithms provided by B. Runow :