Annotated dataset of simulated voiding sound for urine flow estimation

Abstract

Abstract Sound-based uroflowmetry is a non-invasive test emerging as an alternative to standard uroflowmetry, estimating voiding characteristics from the sound generated by urine striking water in a toilet bowl. The lack of labeled flow sound datasets limits research for developing supervised AI algorithms. This work presents a dataset of simulated urinary flow sound recordings at flow rates from 1 to 50 ml/s, in increments of 1 ml/s, against water in a real toilet bowl. Flow generation employed an L600-1F precision peristaltic pump, with simultaneous recordings from three devices: high-quality Ultramic384k microphone, Mi A1 smartphone and Oppo smartwatch. Water was expelled through a 6 mm diameter nozzle (simulating the urethra) from a variable height of 73 to 86 cm, mimicking adult urination. The dataset provides 60-seconds labeled, constant-flow audio recordings (WAV format). This resource is intended to support research on sound-based urinary flow estimation by developing and validating supervised artificial intelligence algorithms.