Subjective perception of laboratory sound anomalies and its relationship with objective evaluations using machine learning

抄録

Studying anomalous sounds in a laboratory is expected to help reduce the frequency of laboratory accidents. This study focused on subjective perception of humans in how anomalous sounds are perceived within the routine laboratory environment and aimed to explore the relationship between subjective perception and objective machine learning evaluations. Two experiments were designed and conducted: one focused on the anomaly perception for various sounds heard in a laboratory, and the other compared the anomaly perception with machine learning assessments. The experimental results identified significant physical and psycho-acoustic features that influenced the participants’ perception of anomalies. Furthermore, we found that the subjective perception and objective evaluation of acoustic anomalies generally agreed when the log-likelihood value was low, but not necessarily when it was high. These findings contribute to the development of real-time monitoring systems for detecting potential anomalies in laboratory environments and offer a novel perspective for laboratory safety management.