We model the audio-scene as a collection of a few dominant sound sources. These sources are assumed to possess stationary properties that can be characterized using a few features. An audio-scene change is said to occur when the majority of the dominant sources in the scene change.

We use three types of features (a) scalar sequences (b) vector sequences and (c) scalar point data. The scalar sequences are further analyzed using:

1. Trends
2. Periodic components
3. Noise

We have developed novel metrics to analyze the different types of features used in our work. Then, these metrics are used to determine (within our memory model framework) a correlation value per one second interval. Local correlation minima give us audio scene changes.