Introduces DCASE 2025 task providing device information at inference for low-complexity acoustic scene classification, with baseline accuracy of 50.72% improving to 51.89% via device-specific fine-tuning.
The most commonly used meth- ods in 2023 and 2024 were augmentation-based methods, such as Freq-MixStyle [7,8] and device impulse response augmentation [9]
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Low-Complexity Acoustic Scene Classification with Device Information in the DCASE 2025 Challenge
Introduces DCASE 2025 task providing device information at inference for low-complexity acoustic scene classification, with baseline accuracy of 50.72% improving to 51.89% via device-specific fine-tuning.