A field experiment on the effect of sound absorption on acclimation of new children to kindergarten—Time ratio of children’s crying by acoustic event detection with neural network

A field experiment was conducted in actual kindergarten to examine that sound absorption help acclimation of new children to kindergarten by reducing bustling atmosphere and improving speech intelligibility. There were two classrooms for two 2-year-old groups with about half of them newly entered the kindergarten. Sound absorbing material was temporarily installed to one of the two rooms reducing the reverberation time from 0.8 to 0.4 s at 1 kHz octave band. The time ratio at which children were crying was chosen as an indicator of their acclimation to kindergarten. Since judgement of crying by human using recorded video and sound requires much labor, detection of crying by acoustic event detection technology was first tried in our previous study. The present study attempted to improve the method using neural network algorithm. A model using CNN was designed and Mel Filter Bank(MFB) and Mel Frequency Cepstral Coefficients(MFCC) were used as features for training and evaluation. Sound clips containing or not containing crying were excerpted from the audio recordings taken in each room to make training data, and test data were taken from the recordings on the other days. Then, the optimal model and features, as well as their setting conditions, were examined.