SMART-Ped
Abstract
SMART-Ped is an open dataset of pediatric respiratory sound recordings acquired using smartphone built-in microphones in a real-world clinical setting. It includes 1500 de-identified respiratory sound recordings from 209 children aged 0–17 years, collected between September 2020 and June 2025 at a tertiary hospital. Recordings were obtained using 15 smartphone models at four auscultation locations (right anterior, right posterior inferior, left posterior inferior, and trachea). All released recordings underwent manual quality assessment prior to publication, and only files meeting predefined quality criteria were retained. Each recording was independently annotated for the presence and type of adventitious sounds, including wheezes and crackles. The repository includes WAV audio files, participant- and recording-level metadata, and predefined five-fold cross-validation splits to support reproducible development and benchmarking of artificial intelligence models for pediatric respiratory sound analysis. This dataset was developed to support research in digital lung auscultation, respiratory audio processing, and pediatric respiratory monitoring. This is an updated version which incorporates minor revisions of the metadata including respiratory sound annotation and the smartphone-related information.
