Datasets

• Positive labled samples Store in ~/breath_data
• Negative samples Store in ~/ESC-50-master

This should result in this directory structure:

root
│   README.md
│   Python scripts
|   requirements.txt
│   
└─── breath_data
│    │   exhale/
│    │   inhale/
│    │   silence/
|    |   ...
│
└─── ESC-50-master
     │   audio/
     │   meta/ 
     │   ...

Usage

Inference

python infer_improved.py 
--weights WEIGHTS 
--out_dir OUT_DIR 
--wav WAV # If unspecified uses microphone
--dropout DROPOUT
--l2 L2
--smooth SMOOTH 
--mode {tile195,slide0975}

Training

python3 train_improved.py
--breath_root BREATH_ROOT # Path to ~/breath_data
--esc50_root ESC50_ROOT 
--cache_dir CACHE_DIR 
--epochs EPOCHS 
--batch_size BATCH_SIZE 
# More flags available, run python train_improved.py to see