Automatic detection of relevant acoustic events in kindergarten noisy environments

In many studies conducted to monitor the health situation of kindergarten child care workers in Germany, the high noise level in the facilities has been pointed out by approximately 70% of the workers as one of the most stressful factors. One factor contributing to the stress is considerable background noise in kindergartens, many important events such as calls for help of children or colleagues might be unheard at their first utterance. This contribution presents results of a study conducted in a real kindergarten for which machine-learning approaches were tested to detect and classify acoustic events in typical background noises. For the training of applied approaches daily Kindergarten noise has been recorded for several weeks and a list of relevant missed events to be annotated from the recordings was provided by child care workers. The goal of this study is to develop an automatic acoustic monitoring system to considerably reduce the number of unrecognized, desired events important for child-care workers