Acoustic scene change detection

This work proposes an acoustic scene change detection approach in the indoor space by reverberation. Speakers emit impulse signals periodicity. Sound waves are then influenced with objects, the floor, walls, and roofs, and generate corresponding reverberation. This reverberation can represent the characteristic of the indoor space, which is called acoustic scene. When the state of the indoor space changes, such as furniture changes positions, windows are open or closed, and the number of people in the room increases or decreases, the acoustic scene would differ from the one before. In this way, we can evaluate whether the indoor scene changes or not. This work presents the difference between the current acoustic scene and the initial acoustic scene by calculating the difference between the current spectrogram and the initial spectrogram of the reverberation. When the sum of acoustic scene differences is greater than the threshold decided in the training stage, we say that acoustic scene change occurs.