Preprocessed Dataset for FAT2019 Multi type processed sound dataset for [Freesound Audio Tagging 2019](https://www.kaggle.com/c/freesound-audio-tagging-2019). Preprocessed Types This dataset contains 3 preprocessing types using full-time range: - Log-Mel-scale Spectrogram 128xN - Log-Mel-scale ΔΔPhase Spectrogram 128xN - MFCC 64xN And, it contains excluded silent sound rows of train curated and noisy CSV. Configure class AugmentationConfig: padding_scale = 1. whitenoise = True whitenoise_level = 1e-3 # 0.~1. pitchshift = True pitchshift_steps = 2. # steps (12/oct) class PreproConfig: gen_mels = True gen_phase = True gen_mfcc = True gen_augmentation = True sr = 44100 duration = 2. # secs n_out = 128 n_mfcc = 64 # generaly 13 n_mels = 128 n_fft = n_mels * 20 fmin = 20 fmax = sr // 2 phase_root = (1. / 2.) hop_len = int(sr * duration // n_out) sample_size = int(sr * duration) padding_size = int(sample_size * AugmentationConfig.padding_scale) Preprocessing Kernel Preparing..