Auto-AR: Automated Situated Design of Augmented Hearing Reality Algorithms

pHearing aid (HA) clients need a personalized and acoustic scene-dependent audio processing algorithm. Therefore, HA algorithms should preferably be designed automatically (i.e., without engineers in the loop) and in-the-field when the environmental conditions and personal expectations can be measured, i.e., under “situated” conditions. Brains provide an inspiring example for automated situated algorithm design. After all, brains design algorithms for speech and object recognition, for locomotion, etc., through situated interactions between the brain and its environment without engineers in the loop. Over the past decade, a large body of theoretical neuroscience work, covered by the umbrella term “the Free Energy Principle” (FEP), has revealed the underlying physical principles of the brain’s automated algorithm design method. The central challenge in this project is to transfer FEP-based design from an academic neuroscientific interest to a robust and practical engineering method, in particular in the context of situated HA design. In this project we will develop a FE minimizing agent that is capable automated situated design of an algorithm that can attenuate or enhance any of the constituent source signals in an acoustic mixture. Moreover, the project will develop an open-source engineering toolbox that can be utilized to specify FE minimizing agents for alternative applications./p