A strong
auditory illusion created when
lip-reading does not match the speech sounds heard. It is normally only salient in specially-prepared filmclips, but may contribute to the occasional unnaturalness of
dubbing.
It was first observed by Harry McGurk (d. 1998) and his assistant John MacDonald, and written up in a paper in Nature in 1976, "Hearing lips and seeing voices" (246, pp. 746-748). They were experimenting on babies. I'm sorry, I'll read that again. They were experimenting on babies' perceptions of their mother's voice, and wanted to know which they would react to if they heard the sound [ba] but saw the lip movements for [ga]. To their surprise, on viewing the clip themselves, they heard neither. They heard [da].
The illusion is persistent and inescapable. Even after you know what's going on, when you watch the clip over and over again you hear [da da da da da], then the instant you close your eyes it becomes [ba ba ba ba ba], then open them and it's back to [da da da]. Almost everyone perceives it this way.
You can learn something about your peripheral vision by seeing what angle you need to turn to to have it switch.
Most of us can't lip-read fully, though even without training we recognize some of the more obvious features: labial consonants like [b p m] involve lip closure, and rounded vowels like [u o] are visibly round-lipped. The sound of [g] is not visually obvious: it's formed primarily by the back of the tongue, and we don't see that, only the secondary effects of tooth and jaw position.
The place of articulation of [d] is mid-way between those of [b] and [g], so it is a compromise sound perceived by the fusion of the two contradictory inputs. However, it is not clear in detail what is going on. It is not immediately apparent from the phonetics that distant sounds should blend or fuse into a physically intermediate one like this. Many West African languages have a labiovelar sound [gb] that sounds like the two simultaneously: why don't we hear this? It's not just that [d] is familiar and [gb] unfamiliar, since trained phoneticians are unable to avoid the McGurk illusion.
In any case, it's not as if there are two sounds fusing. If you actually superimposed a [b] over a [g] you might get the formants or characteristic sound energies to fuse into something more like [d] than [gb], depending on exactly how the sounds line up. But there is only one sound transmitted in the McGurk effect: a pure and simple [b], and with your eyes closed that's all you ever hear.
Quicktime .mov here and elsewhere: http://www.media.uio.no/personer/arntm/McGurk_english.html