Enhancing auditory safety warnings in highway construction zones with loud noise using generative artificial intelligence

Safety-critical sounds are crucial to alert nearby workers of potential dangers such as reversing mobile equipment at construction sites. However, due to the loud ambient noise on highway construction sites, workers may wear hearing protection devices (HPDs), potentially impacting their ability to detect these alarms. There is a lack of existing frameworks that can protect workers from loud and complex noise but preserve the audibility of safety cues. This study addressed this gap by developing a novel framework using a conformed-based metric generative adversarial network (CMGAN) to augment the audibility of safety-critical signals for construction workers exposed to loud noises. The CMGAN-based signal enhancement model was trained on a large-scale synthetic data set of noisy warning signals, covering two types of safety-critical sounds - backup alarms, and work zone intrusion alarms - across 38 background noise conditions such as construction equipment, hand tool operation, and passing traffic. The models were evaluated using different test sets with varying signal-to-noise ratios (SNRs) and objective metrics such as perceptual evaluation of speech quality (PESQ), composite measure for signal distortion (CSIG), composite measure for noise distortion (CBAK), composite measure for overall speech quality (COVL), and segmental signal-to-noise ratio (SSNR). Furthermore, the authors assessed the real-time performance of the proposed models by measuring the real-time factor (RTF). The proposed model shows promise for developing an innovative noise-cancelling device, which could significantly enhance worker safety in construction zones and other high-noise environments.