A comparison of two active‐speaker‐detection methods suitable for usage in noise dosimeter measurements

Measuring noise exposure in a working environment is often done by using standard noise dosimeters. This method is suitable for the evaluation of many working environments. However, in some situations the worker uses his/her voice a large amount during the day, e.g. teachers in a preschool environment. Thus, in these situations regular dosimeter measurements will not correspond to the actual noise exposure. In order to provide correct measurements, methods that can detect when the workers own voice is active are required. This paper presents a study of two such methods originating from voice research; the binaural and the throat microphone methods. The methods are compared using a receiver operating characteristics based method, where the performance is assessed by the Probability‐of‐failure measure, i.e. the percentage of own voice that the method failures to detect correctly. The evaluation is performed in a lab environment as well as in real field conditions in a preschool. The results of the study show that both methods can be successful in a controlled low noise ( 6%).