TR2026-100

The MERL Systems for DCASE 2026 Challenge Task 4

    •  Fujimura, T., Wichern, G., Masuyama, Y., Boeddeker, C., Saijo, K., Richter, J., Edo, T., Le Roux, J., "The MERL Systems for DCASE 2026 Challenge Task 4," Tech. Rep. TR2026-100, IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE Challenge), June 2026.
      BibTeX TR2026-100 PDF
      • @techreport{Fujimura2026jun,
      • author = {{Fujimura, Takuya and Wichern, Gordon and Masuyama, Yoshiki and Boeddeker, Christoph and Saijo, Kohei and Richter, Julius and Edo, Takahiro and Le Roux, Jonathan}},
      • title = {{The MERL Systems for DCASE 2026 Challenge Task 2}},
      • institution = {IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE Challenge)},
      • year = 2026,
      • month = jun,
      • url = {https://www.merl.com/publications/TR2026-100}
      • }
  • MERL Contacts:
  • Research Areas:

    Artificial Intelligence, Machine Learning, Speech & Audio

Abstract:

In this report, we present our anomalous sound detection (ASD) systems for DCASE 2026 Challenge Task 2. Our approach introduces noise-aware audio self-supervised learning (NA-SSL) to leverage two-channel recordings, in which one microphone is used to capture noise. NA-SSL models are trained to extract clean SSL representations from two-channel noisy signals simulated on external datasets. Then, we perform ASD in the extracted denoised feature space. To further improve performance, we perform discrim- inative fine-tuning with attributes and pseudo labels. Furthermore, for anomaly score calculation, we employ several recent techniques: score rescaling, frequency-wise memory bank construction, and deviation-based pooling. Our final ensemble system has achieved 66.20% in the official scores calculated as a harmonic mean of the area under the curve (AUC) and partial AUC (p = 0.1) over all machine types and domains in the development set.