NAASA - Neural Architectures for Acoustical SceneAnalysis


This project aims to develop neurohardware for acoustical scene analysis. The work focusses to attention-driven extraction of auditory information (cocktail-party-effect), and its usage for sensorimotor interaction of robot systems in unknown environments. Different models of acoustical preprocessing, source localization, attention-based source separation and learning of space-time patterns are simulated and tested on real audio signals.

In collaboration with the Department of Microelectronic Circuits and Systems MSS at TU Ilmenau we work toward the VLSI implementation of a spike-based source localization model.