DPI-Watch: An Inhaler Sound Dataset

The DPI-Watch dataset has been collected using a Google Pixel Watch 2 as the audio capture device and placebo DPI inhalers (Symbicort Turbuhaler). Seven adult participants (six males and one female), aged 27 to 47, participated in the data collection. The participants were all researchers based at the Quality of Experience Laboratory (QxLab) at University College Dublin (UCD), Ireland. Each participant wore the smartwatch on their wrist throughout the data collection process. The participants were instructed to use the inhaler correctly based on a protocol, as typically performed in a clinical procedure. They then used the placebo inhaler multiple times based on the given instructions, and at the same time, inhaler sounds were recorded using the smartwatch. There were approximately 10-second pauses between each inhaler use. The data collection was conducted in a quiet room. The dataset consists of 71 PCM WAV recordings, each lasting between 9 and 28 seconds. Each recording captures the administration of a single dose, which was manually segmented into three classes: actuation (device activation), exhalation, and inhalation, based on auditory inspection. This segmentation resulted in 68 actuation, 123 exhalation, and 71 inhalation segments. Some recordings include patient errors, such as missed activation or exhalation, as well as weak or short inhalation. All recordings have a sampling frequency of 44,100 Hz. Please cite the following paper if you use the dataset: Respiratory Inhaler Sound Event Classification Using Self-Supervised Learning