Why Do We Hear Speech Sounds That Noise Has Erased?

A cup suddenly clatters in a café and masks half a sound in your friend's sentence, yet the sentence may still seem continuous and complete. This is not always a later guess about meaning. Under suitable conditions, listeners can experience the missing speech as though it were present. Psychologists call this phonemic restoration.

In Warren's classic 1970 experiment, a speech segment was removed from a sentence and replaced by a noise such as a cough. Listeners often reported an intact word and struggled to locate the interruption. The noise needs to be plausible as something that could have masked the original sound, rather than leaving an obvious silence. The word and sentence context must also support a reasonably likely candidate. The auditory system combines the acoustic input with predictions about language structure to produce the most coherent percept.

This does not mean the ear secretly recreates the original sound wave, nor can every gap be repaired. Restoration weakens when context is ambiguous, the noise could not plausibly mask the sound, the missing stretch is too long, or listening conditions change. Later recordings from human cortex found activity representing masked but perceived speech sounds, although no single experiment shows that the brain uses only one fixed mechanism.

The lesson is more precise than saying that people “fill things in”. Everyday hearing is inferential. If we merely received each acoustic point, coughs, traffic and broken signals would repeatedly collapse ordinary conversation. Context usually improves efficiency, but where several words remain possible it can also steer perception towards the wrong one. Hearing a complete sentence and having a complete recording are therefore two different claims.

https://doi.org/10.1126/science.167.3917.392
https://doi.org/10.1038/ncomms13619