CAA-Net: Conditional Atrous CNNs with Attention

A Python code for the subtask B of the task 1 in DCASE 2018/2019.

Data

DCASE 2018 Task 1 - Acoustic Scene Classification, containing two tasks:

subtask A: data from device A

subtask B: data from device A, B, and C

This code is working on the dataset of subtask B.

Preparation

channels:

• pytorch
• defaults dependencies:
• matplotlib=2.2.2
• numpy=1.14.5
• h5py=2.8.0
• pytorch=0.4.0
• pip:
  • audioread==2.1.6
  • librosa==0.6.1
  • scikit-learn==0.19.1
  • soundfile==0.10.2

File structure

• CAANet_DCASE_ASC
  • pytorch
  • utils-pred
  • runme.sh

Note:

• The folders "pytorch-pred" and "utils-pred" are corresponding to multi-task conditional training.

• The folders "pytorch-wopred" and "utils-wopred" are corresponding to teacher forcing conditional training.

• Please change the folder names as "pytorch" and "utils-pred" before running the code.

Run

sh runme.sh

In runme.sh, please run the following files:

1. feature extracttion: utils/features.py
2. training a model, and evaluation: main_pytorch.py

Cite

If the user referred the code, please cite our paper:

Z. Ren, Q. Kong, J. Han, M. D. Plumbley and B. W. Schuller, "CAA-Net: Conditional Atrous CNNs with Attention for Explainable Device-robust Acoustic Scene Classification," in IEEE Transactions on Multimedia, doi: 10.1109/TMM.2020.3037534.

Zhao Ren

Chair of Embedded Intelligence for Health Care and Wellbeing

University of Augsburg

18.11.2020