Split audio: https://unix.stackexchange.com/questions/280767/how-do-i-split-an-audio-file-into-multiple
Audio IO and using the proposed frame processing methods on audio data: wav_read.py
Load user field data and saved the processed data as class-based: data_preparation_user.py (wav_read.py, check_dir.py)
Load class-based data (option 0 from data_preparation_user.py) and saved the processed data as fold-based: concatenate_class_data.py (wav_read.py, check_dir.py)
Load (processed) user data and do mfcc extraction, 5-fold cross validation: train_test_user.py (wav_read.py)
==============================================================
Enhance web collected audio (format conversion and volumn change): audio_enhancement.py
Calculate Mechanical Turk results: read_mturk_files.py
Test an audio clip with privacy protection: wav_test.py (wav_read.py, pca.py)
==============================================================
Partially adopted from: https://github.com/anuragkr90/weak_feature_extractor, by Kumar et al., ICASSP 18
Load the esc wav files, degrade, and save the degraded mfcc frames(in csv) or full wav clips(in wav): load_esc.py (wav_read.py, check_dirs.py)
Load processed wav, compute segment spectrogram, extract embedding features, and save as csv (require torch): feat_extractor.py (extractor.py, network_architectures.py)
Evaluation for esc features:
for embedding features from transfer leanring: train_test_esc_tl.py (check_dirs.py)
for mfcc features: train_test_esc_mfcc.py (check_dirs.py)