A personalized automated system designed to assign hazardous noise exposures to tasks among agricultural workers

Farming is a noisy occupation, resulting in a high prevalence of hearing loss among agricultural workers. The aim of this study was to improve the accuracy of an automatic algorithm designed to cluster individual sound events into tasks. This work is part of the HearSafe Study, which aimed to increase agricultural workers' use of hearing protection devices by providing personalized information on hazardous noise to workers. Participants in the study interacted with the HearSafe System: a small sound level meter, a website, and an algorithm to associate noise with tasks. They wore the sound level meter that recorded loud (≥ 80 dBA) sound "events," their location, and audio clips. They interacted with the website to view where and when participants were exposed to hazardous noises during the day. To simplify interpretation, an algorithm clustered individual sound events into tasks based on their proximity in time and location. The system's effectiveness hinges on the accuracy of this clustering algorithm. In Phase I, the accuracy was determined using parameters for time between events (2, 5, and 10 min) and distances between tasks (5, 9, and 18 m). In Phase II, the algorithm was refined to account for pauses in work and riding on equipment. Researchers manually clustered events into tasks by listening to the audio clips. Algorithm accuracy was measured as the percentage of events matching the manual clustering. The automating accuracy was improved from 57% with the base algorithm to 87% with the most accurate algorithm (p = 0.02; 10 min between events, 9 m average distance between tasks, and added the condition to combining consecutive tasks that were within 9 m of each other). Increased accuracy in identifying noisy tasks will improve the efficacy of the HearSafe System to communicate when and where use of hearing protection devices are needed among agricultural workers.