Domain Adaptation for Device-Invariant Acoustic Scene Classification
A study published on arXiv evaluates domain adaptation techniques for acoustic scene classification using CNN-based and transformer-based feature representations. Two methods, domain adversarial neural network (DANN) and conditional domain adversarial network (CDAN), were tested under various domain shifts. DANN consistently provided effective adaptation for both feature extractors, while CDAN only worked well with CNN-based extractors. The research highlights the need to tailor adaptation methods to the underlying feature representation. Experiments were conducted on the DCASE 2020 dataset with multiple devices.
Key facts
- Study evaluates DANN and CDAN for acoustic scene classification
- DANN works consistently for CNN and transformer feature extractors
- CDAN only effective for CNN-based extractors
- Experiments use DCASE 2020 dataset with multiple devices
- Domain adaptation methods may need tailoring to feature representation
Entities
Institutions
- arXiv