The Shadow A Sound Casts
A loud sound hides quiet ones for a fifth of a second afterwards, and for a few milliseconds beforehand.
What it is
A loud burst at 1 kHz, and a very short quiet blip you can move around it in time. Drag across for timing, up and down for level. Under the green curve the blip is predicted to be inaudible. Turn the masker off and it reappears.
Drag the blip to the left of zero, so it happens before the masker, and it’s still hidden.
How it works
Day 20 was masking across frequency. This is masking across time, and it has two quite different halves.
Forward masking is the intuitive one. A loud sound leaves the auditory system briefly less sensitive and it takes about 200 ms to recover. The threshold decays roughly linearly in log time, so most of the effect is in the first 20 ms and the tail is long and shallow.
Backward masking shouldn’t happen. A blip that’s already been heard can’t be un-heard by something arriving afterwards, and yet a blip up to about 20 ms before a loud sound gets reported as inaudible.
The usual explanation is that hearing a very short sound isn’t instantaneous. Perception integrates over a window of some tens of milliseconds, and a loud event landing inside that window dominates the whole integration. The blip isn’t erased, it’s never separately resolved.
The two reaches are asymmetric by an order of magnitude, 200 ms forward against 20 ms backward, and there’s a test asserting that ratio directly.
What surprised me
That backward masking is the short one, and that this makes sense in exactly one direction.
Before looking, I’d assumed the causally impossible direction would be a marginal laboratory effect worth a few milliseconds and the causally sensible direction would be the long one. Both true. BUT the reason they differ by ten times isn’t that one is real and the other is an artefact, it’s that they have completely different mechanisms.
Forward masking is recovery from excitation, and recovery takes as long as it takes. Backward masking is a limit on temporal resolution, so its reach can’t exceed the integration window, which is short by definition.
So the asymmetry isn’t a curiosity, it’s a signature. If backward masking lasted 200 ms too, the integration-window explanation would be dead.
What I would do next
Both maskers at once, before and after, and see whether their shadows add or whichever is nearer simply wins.