US8K_AV: Dataset for Environmental Sound Recognition in Embedded Systems for Autonomous Vehicles

The US8K_AV dataset stands for Urbansound8K for Autonomous Vehicles. It is a collection of short environmental real-field recordings, ranging from 1 to 4 seconds in length, with varying sample rates (8 to 192 kHz), mono or stereo, and converted to WAV compressed @ 32-bit float. All samples have been sourced from public real-field recordings available through the Freesound.org project and are provided under the Creative Commons license - Attribution-NonCommercial 4.0 International. This secondary dataset is based on the original Urbansound8K, but with 6 classes ('car_horn’, 'siren', 'dog_bark', 'children_playing', 'background', and 'silence') prearranged into 10 near-balanced folds, ensuring that clips from the same source recording are contained within a single fold. Metadata is provided to outline the procedure for recreating this dataset from its original source at Freesound.org. Additionally, a Jupyter notebook written in Python and a mapping file for fold distribution offer a detailed guide to reproduce the US8K_AV dataset, including merging less relevant classes into a new one and the creation of a new class. Refer to the complete project of environmental sound recognition in embedded systems for autonomous vehicles at GitHub.   The primary purpose of this dataset is to offer a unique combination of environmental sounds tailored specifically for autonomous vehicle applications in smart cities.