# Preprocessed Audio for Freesound Audio Tagging 2019 The competition is a Kernels-only, we'd like to save our time to make it maximize what can be done in kernel running time. This dataset provides files preprocessed as mel-spectrogram. # _New Version 2_ Fixed preprocessing parameter as follows, converted mel-spectrogram has now half length from version 1; Size 128x128 was 1s sound, now 2s in length 128. This change is basically a bug fix, since duration is set to 2s, then 128x128 should have 2s. It would boost performance basically. conf.sampling_rate = 44100 conf.duration = 2 # sec conf.hop_length = 347*conf.duration # <==== updated from `conf.hop_length = 347` conf.fmin = 20 conf.fmax = conf.sampling_rate // 2 conf.n_mels = 128 conf.n_fft = conf.n_mels * 20 # Content Pickled 3 files for all train curated/noisy/test. # Acknowledgements Thank you for Freesound Audio Tagging 2019 organizers and Kaggle staff, and all the competitors!