Introduction to Acoustic event detection technique

Accurately identifying the occurrence of acoustic signatures, like birds chipping, human speaking, etc., from large audio files.

Introduction

Big data provides multiple applications, provided we clear the first and maybe the biggest hurdle of all, identifying interesting patterns or even segregating the meaningful parts from the junk. Ecologist are facing the same problem, with access to hours of recording of forest with all of its elements in its entirety, how can we separate out the important ones, like the sound of a tiger approaching or birds chipping. This calls for an intelligent technique to separate out the garbage which may include background noise of forest like sound of wind blowing, leaves rustling or event distant traffic (in current context). By removing all these, we may able to identify the interesting parts, which categories as acoustic events.

Step 1: Signal Pre-processing

First, lets get the data in a generic format. To start with, we can down sample the data to 22,050 Hz and with 16 bit depth. This means for 1 second of audio data, we have ~22 thousand samples, where each sample can have around 2¹⁶ (65,536) level of samplings or resolution.