Abstract

Source localization and beamforming using microphone arrays are common strategies for speech enhancement and source separation in various applications related to voice capturing. This thesis focuses on their implementation on hardware with real time constraints. Several algorithms are used and the idea is to increase performances as much as possible. Therefore, mathematical and physical optimizations are considered, as well as improvements due to the board’s specificities. Nevertheless, the work focuses more on handling of basic solutions by the hardware features. A hybrid plateform is at disposal; it puts a reprogrammable computation unit and a more classical cpu together. The improvements are significant and the real-time constraints are achieved notwithstanding the fact that hardware capacities and ressources are limited regarding this application that deals with multiple microphone outputs.

Full Text & Additional Material

The full thesis can be downloaded here.