Exploring the speech chain experimentally: English sound identification accuracy is modulated by phonological factors across adult speakers and listeners

Speech perception is influenced by communicator characteristics and linguistic factors. Using a multi-speaker sound identification task, this study explored the relative contributions of speaker and listener variability. 384 triphones were extracted from 21 native English speakers reading a list of phonetically rich words. The words and the resulting triphones were balanced in terms of sound frequency (five groups), context prominence (three tiers), and word stress (levels: monosyllabic, unstressed/stressed polysyllabic). The mean length of the words was 6.0 sounds (range: 3–14). A total of 22 listeners were instructed to type the consonants they identified in audio clips drawn randomly from all the speakers using a Latin square design yielding a total of 16 408 responses. The mixed effects logistic regression model showed identification accuracy was lowered by lower frequency sounds (z = 4.0), less prominent context (z = 5.7), lack of stress (z = 3.7), and number of sounds in the word from which the triphone was extracted (z = 3.3). Speaker intercept standard deviation was 87% greater than that of listeners. The results corroborate a model of sound identification that considers frequency and context prominence. The results suggest that our experimental design is sufficiently powered to further explore the components of the speech chain.