Acoustic Feature Extraction It is very important to derive suitable sound features for the identification of various human and non-human sounds. There are many acoustic features available in literature; they can be roughly grouped into temporal, spectral, parametric and harmonic features [5]. However, it is computational expensive to utilize all these features. In this work, a number of useful features suitable for the targeted acoustic surveillance applications are developed and they will be described in the following sub-sections. Note that a 16 bit A/D converter with 16 kHz sampling is used to digitize the sound. The digitized signal is processed on a frame-by-frame basis with a frame size and frame shift of 512 and 64 samples, respectively.