Rooster Challenge

This repo contains a solution for detecting rooster sounds from audio files.

Instructions

Install requirements:

• pip3 install -r requirements.txt

Rooster Detection

• Run rooster_audio_detector.py file. You can change the weight settings in detector_conf.yaml if you want to run the code with your own training weights.

Training

1. Run download_prepare_dataset.py file to download ESC-50: Environmental Sound Classification dataset. The script will generate an augmented rooster and background sounds to be trained on. More information on dataset can be found from https://github.com/karolpiczak/ESC-50

• For transfer learning purposes, I used resnest model which was also provided good results in Kaggle's bird call classification competition https://www.kaggle.com/c/birdsong-recognition . The weights I used in the training can be downloaded from this link https://drive.google.com/drive/folders/1BrarBYMUH1V4qlMMzU_dT6WQqPhj9J6q?usp=sharing . The weights need to be placed in weights directory.

• There are two modes of training; train_with_birdcall and continue2train . If train_with_birdcall is configured on train_conf.yaml file, then set weights option in train_conf.yaml as resnet_bird_call. If continue2train is chosen for training mode, you can set any of the output weights; best_loss, best_accuracy or final.

2. Run train.py file to start training. Optionally you can adjust training settings(number of epochs, batch_size, ...) in train_conf.yaml file.