Skip to content

Latest commit

 

History

History
32 lines (29 loc) · 1.78 KB

README.md

File metadata and controls

32 lines (29 loc) · 1.78 KB

SERP Project Bird Audio Detection (BAD) with novel PCEN audio filter

This github repository is a complement to the presenting poster at the NYCSEF 2020 fair. This repository includes all source code of implementing PCEN onto Thomas Grill's Bulbul Deep Learning Model, as well as the data analysis of results and graph production. It had achieved a preview AUC (area under the curve, see poster for details) score of .885 in the DCASE 2018 Bird Audio Detection Challenge.

Results

Application of PCEN was significantly beneficial to the model's performance, from a mean AUC score (out of 5 trials) of .848 to a .904 in this experiment.

Trial 1 Trial 2 Trial 3 Trial 4 Trial 5 Average P-Value
no pcen 0.859158 0.854338 0.820336 0.880204 0.826951 0.848197
pcen 0.914403 0.901837 0.899928 0.901896 0.903068 0.904226
T-Test 0.001097

Prerequisites for code

All components of the project is run on Python 3 (version should not make a difference). Packages used include:

  • Pydub
  • Librosa
  • tqdm
  • h5py
  • Anaconda
    • Numpy
    • Pandas
    • SciPy
    • Scikit-Learn
  • For prerequisites in running the Bulbul model, look here.

Acknowledgements

I would like to thank Dr. Michael I Mandel from Brooklyn College CUNY as well as Dr. John Davis from Staten Island Technical High School for assisting, advising, and supervising me throughout my project.