CP-JKU Submissions to DCASE?20: Low-Complexity Cross-Device Acoustic Scene Classification with RF-Regularized CNNs
Sprache des Titels:
Detection and Classification of Acoustic Scenes and Events Challenge 2020
This technical report describes the CP-JKU team?s submission forTask 1 ? Subtask A (Acoustic Scene Classification with MultipleDevices) and Subtask B (Low-Complexity Acoustic Scene Classifi-cation) of the DCASE-2020 challenge . For Subtask 1.A, we pro-vide ourReceptive Field (RF) regularized CNNmodel as a baseline,and additionally explore the use of two different domain adaptationobjectives in the form of theMaximum Mean Discrepancy (MMD)and theSliced Wasserstein Distance (SWD). For Subtask 1.B, weinvestigate different parameter reduction methods such asPruning,while maintaining the receptive field of the networks. Additionally,we incorporate a decomposed convolutional layer that reduces thenumber of non-zero parameters in our models while only slightlydecreasing the accuracy, compared to the full-parameter baseline.