/DCASE2019-Task1

Code for DCASE 2019 Task 1a, 1b and 1c

Primary LanguageJupyter NotebookMIT LicenseMIT

DCASE2019-Task1

Code for DCASE 2019 Task 1a, 1b and 1c associated with Technical Report:

McDonnell and Gao: "Acoustic scene classification using deep residual networks with late fusion of separated high and low frequency paths", submitted to Task 1 of the 2019 DCASE Challenge (http://dcase.community/challenge2019/task-acoustic-scene-classification)

Inference:

  1. View the inference jupyter notebooks to see confusion matrices from the technical report.
  2. Rerun after downloading trained models (trained on the 70% DCASE training split of the development sets): https://www.dropbox.com/sh/47s97kaw4p8n31g/AAAeWHgl0euewj1ge577_uE9a?dl=0

Training:

  1. Download the data from links at: http://dcase.community/challenge2019/task-acoustic-scene-classification
  2. Adjust the paths in the DCASE_training.ipynb jupyter notebook
  3. Run the DCASE_training.ipynb jupyter notebook