Landmark based audio synchronization: Time synchronization of environmental audio recordings

Abstract

Using different devices to record audio simultaneously usually results in a situation where the recordings may have started at different times. Finding how the different signals align in time can be done by using audio landmarks, which are constructed from peaks in the spectrogram. There are several readily available implementations based on this approach. However, the existing tools are more optimized for music, and usually their main purpose is to find a matching audio track from a database instead.

This work presents a simple algorithm for environmental audio time synchronization. Comparison shows that in case of environmental audio the presented method is faster and more reliable than the existing algorithms.