Video Fingerprinting
One of the most common use cases for audio fingerprinting is TV broadcast monitoring. Generating detailed reports about aired TV content is done by analyzing its audio stream. If you want to receive accurate statistics about the occurrence of a specific ad, applying audio fingerprinting may seem sufficient. And indeed, it works in most cases.
At times though, advertisers use the same audio stream paired with modified visual content. In the automotive and healthcare industries, it is actually widespread. The visual content changes can be minor and sometimes even unnoticeable for the regular viewer. However, it can play a big difference if you want to extract more information about the advertiser, the company, or your competitor.
Let’s start with a couple of examples to outline the problem. Below is an example of a Toyota ad that uses the same audio stream paired with a different video.
The difference starts at the 10th second when the commercial offer appears on screen. It stays on screen for 10 seconds, getting narrated to the audience. Since advertisements are usually read fast and with the same tone of voice, audio fingerprints algorithms may fail to distinguish them.
