Sound Event Detection: A Tutorial

Imagine standing on a street corner in the city. With youreyes closed, you can hear and recognize a succession ofsounds: cars passing by, people speaking, their footsteps whenthey walk by, and the continuously falling rain. Recognitionof all these sounds and interpretation of the perceived sceneas a city street soundscape comes naturally to humans. It is,however, the result of years of “training”: encountering andlearning associations between the vast variety of sounds ineveryday life, the sources producing these sounds, and thenames given to them.Our everyday environment consists of many sound sourcesthat create a complex mixture audio signal. Human auditoryperception is highly specialized in segregating the soundsources and directing attention to the sound source of interest.This phenomenon is called cocktail party effect, as an analogyto being able to focus on a single conversation in a noisy room.Perception groups the spectro-temporal information in acousticsignals into auditory objects such that sounds or groups ofsounds are perceived as a coherent whole [1]. This determinesfor example a complex sequence of sounds to be perceived as asingle sound event instance, be it “bird singing” or “footsteps”.The goal of automatic sound event detection (SED) methodsis to recognize what is happening in an audio signal and whenit is happening. In practice, the goal is to recognize at whattemporal instances different sounds are active within an audiosignal