The library it's part of a larger research project called PiaffNet. Our goal is to make it easier for everyone to study and identify birds by their vocalizations using modern Deep Learning algorithms. PiaffNet uses modern standard of implementations in Python language. We priveledge simplicity and good enough accuracy over unnescessory complexity.
The birdsong classifier library provides a simple predictive model based on supervised learning over data samples of annotated bird sounds. We use a Convolutional Neural Network (CNN) models. Note that this version implemented in TensorFlow library.
The classifier uses a MEL sounds spectograms with
pip install --upgrade pip
pip install -r requirements.txt
pip list
When in project directory
mkdir raw_data
mkdir raw_data/train_audio
mkdir raw_data/split_data
mkdir raw_data/images_png
Copy your files in the train_audio directory, the program uses audio_dataset_from_directory so it needs to be in this format :
train_audio/
species1/
song1.audio
song2.audio
...
species2/
song1.audio
song2.audio
...
...
Slice your audio and detect silent segments :
make run_slicing
make run_preprocess
make run_train
make run_predict