Skip to content

Latest commit

 

History

History
69 lines (53 loc) · 2.93 KB

README.md

File metadata and controls

69 lines (53 loc) · 2.93 KB

Bird Audio Detection Challenge 2018 - DCASE Task 3

This is the submission for DCASE 2018 Task 3 by UKYSpeechLab

Required Software:

  • Python version 3.5
  • Tensorflow-gpu (tested on v1.8)
  • Keras
  • other packages and dependencies listed in requirements.txt

Project directory structure

For running this project on the DCASE challenge 2018 data, follow this directory structure: (unzip the provided compressed files in the main project directory)

  • < project directory >/adaptation_files
  • < project directory >/labels
  • < project directory >/prediction
  • < project directory >/trained_model
  • < project directory >/workingfiles

The directory 'workingfiles' has four subdirectories that contain feature files.

  • < project directory >/workingfiles/features_baseline (download link)
  • < project directory >/workingfiles/features_high_frequency (download link)
  • < project directory >/workingfiles/features_high_temporal (download link)
  • < project directory >/workingfiles/features_ht_enhanced (download link)

In each of these four directories, place the feature files from the provided download links such that each directory should have following six sub-directories:

  • "BirdVox-DCASE-20k" (20,000 files)
  • "Chernobyl" (6620 files)
  • "ff1010bird" (7690 files)
  • "PolandNFC" (4000 files)
  • "warblrb10k" (8000 files)
  • "warblrb10k-eval" (2000 files)

In order to reproduce the results of this submission, place the python files in the main project directory and run in the following order:

  • birddet_baseline.py
  • birddet_high_temporal.py
  • birddet_high_frequency.py
  • birddet_adaptation.py
  • birddet_enhancement.py
  • birddet_multimodel.py
  • computeroc.py

Six prediction files will be generated in the "prediction" directory. These are csv format files having the required format of DCASE challenge submission.

Average of above six prediction files:

Our top scoring submission to the DCASE challenge is a simple unweighted average of the above six predictions. Run the python code "compute_ave.py" from the main project directory after all of the following six prediction files have been generated.

  • DCASE_submission_baseline.csv
  • DCASE_submission_high_frequency.csv
  • DCASE_submission_high_temporal.csv
  • DCASE_submission_adaptation.csv
  • DCASE_submission_enhancement.csv
  • DCASE_submission_multimodel.csv
  • Compute_ave.py will generate "DCASE_submission_final.csv" which is the final prediction file.

Team Members Michael T. Johnson, Narjes Bozorg, Sidrah Liaqat, Neenu Jose, Patrick Conrey, Anthony Tamasi