The script generates a known waveform, it is passed through the browser's audio implementation and then the script analyses the result after. Based on your devices settings and hardware the output will be different, e.g. a PC with analog output might have 44KHz audio output bandwidth, but a bluetooth headset might have a lower, different audio bandwidth. That is a datapoint that can be used in device fingerprinting alongside screen and viewport dimensions,
device pixel ratio, browser plugins, etc.
On the one hand, I wouldn't expect too many variations here (the vast majority of devices probably use 48 kHz and 24-bit output, a few use 44.1 kHz and/or 16-bit, etc.) but just like DPR and all the other properties with a very small set of popular values in practice, you only need a bit or two from each measurement to eventually have a high quality fingerprint.
The script generates a known waveform, it is passed through the browser's audio implementation and then the script analyses the result after. Based on your devices settings and hardware the output will be different, e.g. a PC with analog output might have 44KHz audio output bandwidth, but a bluetooth headset might have a lower, different audio bandwidth. That is a datapoint that can be used in device fingerprinting alongside screen and viewport dimensions, device pixel ratio, browser plugins, etc.
On the one hand, I wouldn't expect too many variations here (the vast majority of devices probably use 48 kHz and 24-bit output, a few use 44.1 kHz and/or 16-bit, etc.) but just like DPR and all the other properties with a very small set of popular values in practice, you only need a bit or two from each measurement to eventually have a high quality fingerprint.