Effects of word context on formant discrimination

Abstract

Studies of phonetic variation identify systematic, but often very small, acoustic differences in the realization of vowels. However, it is not known whether such differences are perceptually relevant in realistic contexts. We examine perceptibility of incrementally manipulated differences in F1 and F2 in vowels embedded in real words of English (experiment 1), and vowels in isolation (experiment 2), via an AX discrimination task. Forty listeners in each experiment, recruited through Amazon Mechanical Turk, heard 27 monosyllabic words or vowels in pairs containing a no-change token and a token edited to differ up to ±100 Hz in F1 or ±150 Hz in F2 relative to the original and were asked to say whether tokens were same or different. Discrimination increased with bigger vowel differences (both F1 and F2) in both experiments; however, even maximally different tokens were correctly discriminated less than half the time. F2 discrimination was better in isolated vowels (48%) than in whole words (33%). Compared to previous results in optimized difference limen studies (e.g., Kewley-Port and Watson, 1994), our results show that untrained listeners in more typical listening conditions (non-synthetic speech heard in non-laboratory conditions in real word contexts) require much bigger differences to be perceptible.