Recent advances of Neural Networks are a game changer in nearly every field of computing, and the acoustics domain is no exception to this. Every branch of acoustics can - one way or the other - benefit from the new methods of analyzing, modifying or generating acoustic data. Such successes make it easy to forget about the nonfunctional side of things: Am I allowed to transfer data to a cloud vendor for processing? Am I allowed to use and share pre-trained models under the EU legislation? How reliable is my system and how susceptible is it for possible attacks? This paper gives an overview of common issues of machine learning applications regarding trustworthiness of the system and privacy of the data. Technical means to overcome them are discussed and assessed regarding their adequacy for topics of concern for the DEGA members.
ICS file for iCal / Outlook