Calculation of activity noise levels in classrooms by using a Gaussian mixture model

The active learning classroom is a new learning space developed to facilitate students' engagement during the class. However, its acoustical conditions, especially in an occupied condition, have not been thoroughly studied. This study investigates the activity noise levels in occupied conditions through unsupervised learning methods. Three clustering algorithms, including K-mean clustering, Gaussian mixture model, and Spectral clustering algorithms, are employed to analyze the continuous 1/3rd octave-band noise levels. The noise data were collected from 5 active learning classes in 3 active learning spaces at Concordia University in Montreal. The algorithms predict the acoustic levels of the lecturer's speech, students' individual speech, student's individual work, student group work, and media presentation. The results obtained from the analysis are compared with the actual results noted by the researcher during the measurements, and the algorithms' performances are evaluated subsequently. Finally, the advantages and disadvantages of using each algorithm are discussed.