Speakers coarticulate less when facing real and imagined communicative difficulties: An analysis of read and spontaneous speech from the LUCID corpus
(3 minutes introduction)
| Zhe-chen Guo (University of Texas at Austin, USA), Rajka Smiljanic (University of Texas at Austin, USA) |
|---|
This study investigated coarticulation of read and spontaneous speech in different communicative contexts from the LUCID corpus. Spontaneous speech samples were from Southern British English speakers who completed an interactive spot-the-differences task with no communicative barrier (NB), with their voice vocoded (VOC), and with a partner who heard their speech in babble (BABBLE) or was a non-native English speaker (L2). The same speakers also read sentences in a casual (READ-CO) and clear (READ-CL) speaking style. Tokens of a pre-defined set of keywords were extracted from the speech samples and consonant-vowel sequences in these tokens were analyzed using a whole-spectrum measure of coarticulation. Results showed that coarticulatory resistance in the six communicative contexts from highest to lowest was: BABBLE > VOC, L2, READ-CL > NB, READ-CO. Thus, in response to communicative barriers, be they real or imaginary, speakers coarticulated less, in line with the models of targeted speaker adaptations (the H&H theory [1] and Adaptive Speaker Framework [2]).
