Audio fingerprint

Also called: AudioContext fingerprint, audio fingerprinting.

Audio fingerprinting never plays a sound. A page builds an offline audio context, runs a synthesized waveform through a processing node such as a dynamics compressor, and reads the resulting sample buffer back as numbers. Those numbers differ in their last decimal places between machines, because the audio stack's floating-point maths is affected by the CPU, the build of the browser and the platform's own libraries. Hash the buffer and you have another stable identifier, collected in milliseconds with no permission and nothing audible.

It is worth knowing about mainly because it is one of the surfaces people forget. A setup that handles canvas and WebGL carefully and leaves audio reporting the host machine's real value has left a join between every profile on that computer — the two treated profiles look different everywhere except the one place nobody checked. Argus treats audio as a first-class surface alongside canvas and WebGL, applied in the renderer and configurable in the same four modes, so it is covered by default rather than by remembering. The mechanics are on the fingerprints page.