DCASE 2020 Task 2
Unsupervised anomalous machine sound detection for condition monitoring.
Developed models for detecting anomalous machine sounds using unsupervised learning for condition monitoring.
Implemented advanced architectures (e.g., NoisyArcMix, SpecNet) and applied feature normalization and model ensembling, improving AUC by 2.14%.
t-SNE visualization of learned embeddings for the fan machine type, before (left) and after (right) applying our feature refinement — machine IDs separate into clearly distinct clusters.