The perception of gated Dutch diphones

The results of a large-scale speech perception study are reported. Eighteen Dutch listeners identified gated fragments of 1179 Dutch diphones. Diphones were presented six times (for some diphones beginning with voiceless stops, four times) to each listener, in fragments ranging from the shortest gate (the first sixth of the diphone), through to the longest gate (the complete diphone). Order of presentation of all gates of all diphones was fully randomized. Listeners were asked to identify the complete diphone on every trial. The results showed that listeners based their decisions on the acoustic information available in the stimuli, not on higher-order factors such as phoneme occurrence frequencies or transitional probabilities. Perceptual confusions reflected the temporal evolution of phoneme similarities (for example, in their shorter gates, long vowels were confused with their short counterparts). This database provides detailed information about the perceptual hypotheses which listeners entertain about phoneme sequences, as those sequences unfold over time. It can also be used to generate predictions about the temporal pattern of activation of spoken words. Such data are necessary for the evaluation and further development of current models of spoken word perception, in which fine-grained acoustic information is mapped continuously onto the mental lexicon.