> I'm assuming "playing silent audio" is a sufficiently common thing for websites to do to have motivated browsers into doing the slightly more complicated thin

> I'm assuming "playing silent audio" is a sufficiently common thing for websites to do to have motivated browsers into doing the slightly more complicated thing of actually analyzing audio streams for content When most sites do browser fingerprinting using audio, they use an 'OfflineAudioContext' - which unlike a normal AudioContext, doesn't run in real time and send its output to system's loudspeakers; instead, it runs faster-than-real-time and outputs to an AudioBuffer. So the browser can easily tell no sound is being played. For example - you can tell the browser to create a 10kHz square wave, feed it to a GainNode then a DynamicsCompressorNode, output to a buffer, then hash the contents of that buffer - which gives you a browser fingerprint, because of subtle differences between browsers.