DCASE Task 2 provides a longitudinal view of how unsupervised anomalous sound detection (ASD) has evolved with changes in domain assumptions, machine coverage, and acoustic information. This focused review compares the official 2025 and 2026 settings and synthesizes 14 primary studies selected under explicit criteria. It organizes the methods by acoustic representation, learning objective, anomaly scoring, domain and noise handling, and post-processing. Two year-separated, singleseed local baseline cases provide descriptive empirical illustrations. In the preserved 2025 evaluation, selective Mahalanobis yielded higher harmonic-mean summaries for source and target AUC, whereas the pAUC summary changed only marginally. In the 2026 local self-contained implementation, it yielded higher targetdomain AUC for five of seven machines and higher section-level pAUC for four of seven machines. These directional patterns motivate machine-level interpretation within each metric and yearspecific protocol. Because the release, dataset, implementation, evaluator, and repeated-run records are incomplete, the findings remain local observations rather than official reproductions or a cross-year benchmark. The comparison also identifies distinct design and reporting questions raised by the 2025 supplementary recordings and the 2026 synchronized near/far channels.
 

anomalous sound detection,DCASE Task 2,machine learing,domain generalization,machine condition monitoring