Perception of all English Sound Sequences: The Diphones Project
Natasha Warner, University of Arizona
collaborators: Anne Cutler, James McQueen, Seongjin Park, Priscilla Shin, Maureen Hoffmann

[Background reading: Warner, McQueen, & Cutler (2014)]

Most speech perception experiments test perception of specific sequences of speech sounds in order to test specific hypotheses. In this study, we tested perception of the 2288 possible two-sound sequences (diphones) of English, such as /an, tʃɛ, pt, oʲæ/ as well as the more usual /ba, ab/ etc. For each diphone, we created six gates, with end points at thirds of each phoneme (e.g. one-third through /a/, two-thirds through it, at the end of /a/, one-third through /n/ in /an/, etc.). Listeners heard each gated stimulus (a total of 13,464 stimuli) and were asked to respond with what two sounds they heard or heard the beginning of. The total dataset comprises over 500,000 perceptual judgments. This allows us to ask questions about how American English listeners use acoustic cues as they become available over time for all possible combinations of sounds. A complementary study on Dutch is available from Warner, Smits, McQueen, and Cutler’s previous work. We have made this data publicly available. In the current talk, I will address several effects on perception, such as the effect of vowel stress, the effect of phonological environment, and the effect of segment probability. The perceptual data also forms the input to the Shortlist-B model of spoken word recognition.