I'm not 100% sure but the original link says
The audio processing did not record people through their microphones. Instead, it generated and processed an inaudible signal, then measured small, repeatable differences in the way a browser and device handled it.
Oh! They have a link to this -> https://ritter.vg/blog-webaudio_alibaba.html
Which makes it sound less scary, and not very effective against FF with fingerprint protection turned on. FF is working on mitigating it. Maybe Chrome and Safari can protect too. But it's also not quite a nothing-burger. Even a weak signal, is useful b/c lots of weak signals combine into a strong signal.
Everything contributes to the fingerprint. Even if everyone falls into say 2 buckets, that's still one bit you didn't have before.