Acoustic data for the manuscript entitled 'Self-supervised acoustic leakage detection for water distribution systems: A real-time diagnosis framework under data scarcity'

This dataset contains 1000 acoustic data mostly collected from an outdoor leakage detection training base located at Dongguan in Southern China. These data have been used to train deep learning based leakage detection models for water distribution systems.This dataset is categorized into three folders. The first is the "leak" folder, which contains 500 one-second audio clips with detailed leak labels. Labels are named according to the following rule: Pipe Material-Region-Pressure in MPa-Flow Rate in m/s-Collection Device, where missing information is denoted as NA. The "no leak" folder contains 386 one-second audio clips with detailed no-leak labels, which follows the same naming rule as the "leak" folder. All data in these two folders were collected on-site from the leakage detection training base mentioned above. There is also an "environmental noise" folder containing 114 one-second audio clips with detailed no-leak labels, named according to the following rule: Noise Category-Region-Collection Device, with missing information also denoted as NA. (Most of the "environmental noise" data were downloaded from the public website (URL: https://www.dlmeasure.com/), while a small portion of the "environmental noise" data were collected on-site from the leakage detection training base.