Assessing the acoustic noise in intensive care units via deep learning technique

Intensive care unit (ICU) noise is a critical and often overlooked issue, impacting patient recovery and healthcare staff well-being. Existing research primarily relies on costly sound level meters for monitoring noise levels, where the characteristics of noise sources cannot be determined and discriminated. This study employs deep neural networks to detect and classify ICU noise events, enhancing source identification. A cost-effective internet of things-based audio recording and monitoring system has been designed and deployed in three ICUs for data collection. The acoustic event classification system described in the paper integrates convolutional neural networks for event detection, followed by clustering to isolate noise sources. Results demonstrate precise classification, with speech identified as a major contributor in all ICUs. This model offers valuable insights for characterising acoustic sources in typical ICUs, which could be the first step towards tackling the problem of excessive noise in ICUs as well as a starting point for further research in this area.