Joint Feature and Output Distillation for Low-complexity Acoustic Scene Classification
Abstract
This report presents a dual-level knowledge distillation framework with multi-teacher guidance for low-complexity acoustic scene classification (ASC) in DCASE2025 Task 1. We propose a distillation strategy that jointly transfers both soft logits and intermediate feature representations. Specifically, we pre-trained PaSST and CP-ResNet models as teacher models. Logits from teachers are averaged to generate soft targets, while one CP-ResNet is selected for feature-level distillation. This enables the compact student model (CP-Mobile) to capture both semantic distribution and structural information from teacher guidance. Experiments on the TAU Urban Acoustic Scenes 2022 Mobile dataset (development set) demonstrate that our submitted systems achieve up to 59.30\% accuracy.
Cite
@article{arxiv.2507.19557,
title = {Joint Feature and Output Distillation for Low-complexity Acoustic Scene Classification},
author = {Haowen Li and Ziyi Yang and Mou Wang and Ee-Leng Tan and Junwei Yeow and Santi Peksi and Woon-Seng Gan},
journal= {arXiv preprint arXiv:2507.19557},
year = {2025}
}
Comments
4 pages, submitted to DCASE2025 Challenge Task 1