Perception of coda voicing from properties of the onset and nucleus of 'led' and 'let'

Syllable-onset /l/ in British English is longer and often has different (usually lower) F2 frequency before a voiced coda. Five experiments explore the perceptual power of these properties and of f0. In each experiment, listeners identified as led or let synthetic syllables whose latter half was replaced by noise. The most reliable cue was /l/ duration; F2 frequency in the /l/ was influential mainly when the vowel quality was held constant. However, listeners learn which cues are most effective, and some choose /l/ duration rather than spectral properties relatively late in the procedure. The results support word recognition models with non-segmental lexical representation that is sensitive to systematic variation in phonetic fine detail.