DUMMY PROTOTYPICAL NETWORKS FOR FEW-SHOT OPEN-SET KEYWORD SPOTTING: abstract simplified (18062976)

In this abstract, the authors describe a system and technique for processing audio data using a dummy prototypical network. The system is designed to perform few-shot open-set keyword spotting (FSOS-KWS). The process involves determining prototype representations based on support samples associated with different classes. Each prototype representation is associated with a specific class. Additionally, a dummy prototype representation is determined in the same learned metric space as the prototype representations. Distance metrics are then calculated for query samples based on the prototype representations and the dummy prototype representation. These distance metrics are used to classify each query sample into one of the classes associated with the prototype representations or into an open-set class associated with the dummy prototype representation.