Table 1 Dataset release specifications.
From: A dataset for environmental sound recognition in embedded systems for autonomous vehicles
Topic | Description |
|---|---|
Subject | Computer and Information Sciences |
Specific subject area | Recognition of environmental sounds in the context of autonomous vehicles using embedded systems for smart cities. |
Type of data | Audio (processed from online source). Additionally, a table (.CSV format) describing the audio metadata from the online source. |
Data collection | The primary data comes from the dataset US8K. Irrelevant classes related to the purpose of the final dataset were merged into a new class keeping the same balance ratio from the primary dataset. An additional class was incorporated using audio samples sourced online, adhering to the same methodology as the original data collection (audio must be real-field recordings). The resultant dataset comprises 4,908 WAV files, totaling 4.94 hours of annotated audio samples, which are distributed among 6 classes and partitioned into 10 folds. Special attention was given during the partitioning process to prevent data leakage from audio samples originating from the same online source. |
Data source location | Online source (Freesound.org) of real-field recordings around the world. |
Data accessibility | Dataset is hosted in the Harvard Dataverse. |
Repository name: Harvard Dataverse. | |
Data identification number: 10.7910/DVN/4D8WPK | |
Direct URL to data: https://doi.org/10.7910/DVN/4D8WPK | |
Public access is granted by consenting to the usual License/Data Use Agreement and Terms of Use (CC-BY-NC 4.0) |
