/BirdCNN

A predictive model to identify four species of bird from their vocalisations. It is my contribution to a Hackster competition.

Primary LanguagePython

BirdCNN

This repository is structured as follows.

Train

The train section uses TensorFlow to train a Convolutional Neural Network that can identify four different bird species by their distinct calls. These birds are:

  • Barn Owl
  • Crow
  • Oriental Scops Owl
  • Western Screech Owl

Predict

The predict section uses the model produced in train to make predictions on new audio recordings of bird sounds passed to it. Any new recordings are expected to be preprocessed using the supplied preprocess.sh script beforehand.

Dependencies

  • Python 3
  • Tensorflow
  • ffmpeg
  • Numpy
  • sox

Credit

Data used in train is a collection of primary and secondary data available in the repository here. Secondary data is provided by Macaulay Library at the Cornell Lab of Ornithology, with all used data credited in their respective directories of the repository linked above.