We do not perceive just speech sounds, we perceive individual speakers' characteristics : indexical properties in speech processing

The traditional approach to speech perception has relied on the assumption that speech is structured in systematic ways and that the linguistic information encoded in the speech signal can be represented reliably and economically as a sequence of abstract, linear units. Speech has been thought to be “basically a sequence of discrete elements” (Licklider 1952: 590), for “in writing we perform the kind of symbolization (…) while in reading aloud we execute the inverse of this operation: that is, we go from a discrete symbolization to a continuous acoustic signal” (Halle 1956: 510). The word feel is traditionally represented as composed of three segments /f/ /i:/ /l/, sequenced in a linear fashion. It is differentiated from the word veal /vi:l/ by the feature of voicing, with /f/ being voiceless and /v/ being voiced. Segmental representations are thus designed to code only the linguistically significant differences in meaning between minimal pairs of words in the language (Twaddell 1952)