A new study shows that attribute-aware pre-training with pseudo-attribute labels boosts anomalous sound detection, achieving higher accuracy on the DCASE 2025 Challenge dataset. https://getnews.me/attribute-aware-pre-training-boosts-anomalous-sound-detection/ #anomaloussounddetection #dcase2025

One‑stage supervised contrastive learning (OS‑SCL) reached 94.64% AUC on DCASE 2020, rising to 95.71% with the new TFgram feature. Code is available on GitHub. Read more: https://getnews.me/supervised-contrastive-learning-improves-anomalous-sound-detection/ #anomaloussounddetection #tfgram

An Enhanced Audio Feature Tailored for Anomalous Sound Detection Based
on Pre-trained Models
Guirui Zhong, Jun Du et al.
Paper
Details
#AnomalousSoundDetection #PretrainedModels #AudioFeatureEnhancement