JGrijalba/Data used for the automatic classification of the physical composition of the urban soundscape

Abstract

This repository documents a manuscript on urban soundscape composition. It contains supplementary materials, supporting data used to train and deploy a convolutional neural network (CNN) for soundscape classification in Venice, Italy (U-VNE-CNN), and data required to reproduce the results presented in the manuscript. The repository includes: U_VNE_original_audio.zip: Original audio data used for labelling and mel-spectrogram generation (org_audio_pb). U_VNE_augmented_audio.zip: Augmented audio segments (aug_audio_pb). U_VNE_mel_spectrograms.zip: Mel-spectrograms organised into five folders: (1) all original mel-spectrograms (org_mels_pb); (2) augmented mel-spectrograms from the training split only (aug_mels_pb); (3) all original mel-spectrograms combined with the augmented training mel-spectrograms (org_aug_mels); (4) training mel-spectrograms, including augmented mels, combined with validation mel-spectrograms (train_val_mels_pb); and (5) test mel-spectrograms (test_mels_pb). U_VNE_mel_spectrograms_dep.zip: Mel-spectrograms generated from audio segments recorded at all sampling points and periods used for CNN deployment (VNE_mels). U_VNE_maps.zip: Geographic data used to calculate morphological indicators and generate soundscape maps (GIS). Supplementary_Materials.pdf: Additional materials and analyses related to the accompanying manuscript. Contact Julián Grijalba at julian.grijalba@rai.usc.es for questions related to this repository or if you have an interest in the original *.wav recordings of Venice (Italy). Please note that the underlying software, particularly the CNN implementation, may become unstable as Python libraries are updated.