Analysis of Glass Breaking Sound with High Density Spectral Data

Glass Breaking is a sign of trouble. It can represent a fight, burglary, a breakin, or an accident, all requiring further action to handle it. So a "correct detection of a breaking glass” is always a need from security.

Glass Breaking Sound is so special that humans seldom mix it up with other sounds. However, in conventional sound detection technology, it would easily mix up with clapping sound. Are those sounds look alike in sound analysis?

With our analysis tool, we can take a look under the high density spectral data view of Glass Breaking Sound and Clapping Sound. Fig.1 is with two clapping sounds, showing a burst of sound with several frequency components ranging from 300~2000 Hz, then a quick drop in power to a stop.

In Fig.2 is the sound of a piece of glass breaking and shattered into pieces. Other than those components with a burst of sound, there are many other components with lesser power but similar patterns, also the power drop rate is steeper in glass breaking. And there are high power components with gentle slopes in the power drop at the same time.?

Conventional sound analysis is unable to detect a power drop of a specific component, making it only able to detect a glass breaking as a high power burst within a short time. That is a common characteristic for both clapping and glass breaking sound. This causes the sound detection to mix those together.?

With our tool, which is able to trace the power change of each component, is able to distinguish glass breaking sound accurately. This also enables us to extract more information in the breaking event, like if it shattered? Anyone stepping on the broken glass? or just hitting the glass without breaking it? All of those can be found with the detail changes in sound.