My primary research interests are in the realms of phonetics, psycholinguistics, and computational linguistics. My current research program is focused on the variation in speech, specifically looking at how listeners accommodate speaker variation and how variable the realization of a word can be across conversational speech. My dissertation research focuses on the listener side of variation, using manipulations within a perceptual learning paradigm.
Courses Currently Teaching
Phonological CorpusTools (PCT) is a tool for inspecting corpora and quantifying aspects of the phonological system, such as the predictability of distribution or the frequency of alternation for two segments. Future releases will support calculating psycholinguistic measures across corpora, such as neighbourhood density and phonotactic probability, and allow for acoustic and corpus analysis of spontaneous speech corpora.
Exemplar Network Explorer
Exemplar Network Explorer (ENE) is a tool for clustering acoustic tokens and analyzing the network that results from this clustering. ENE is currently under active development, and source code may be found on Github.
Variation in speech perception
My dissertation, supervised by Molly Babel, focuses on how listeners accommodate highly variable input. Using a perceptual learning paradigm, listeners are exposed to a speaker with ambiguous productions of /s/. Previous research has found that following such exposure, listeners expand their category of /s/ at the expense of /sh/. However, the expansion is implied in the literature to rely on linking the ambiguous production with the category, which in turn requires that the listener be tolerant of variation. The manipulations in this experiment focus on that tolerance, either increasing it or decreasing it. Through these manipulations, I explicitly test the underlying assumption that ambiguous sounds must be unambiguously associated with a category to affect a listener’s perceptual system, and mere exposure to them is not sufficient.
Variation in speech production
I have worked extensively with the Buckeye Corpus, looking at segmental durations, vowel dynamics (preceedings of InterSpeech 2012), and holistic reduction measures (poster presentation at LabPhon 2014). The vowel dynamics work used smoothed-spline ANOVA (SSANOVA) to assess trajectories of vowels, and the holistic reduction measure used clustering techniques on distances of mel frequency cepstrum coefficient (MFCC) representations.
Acoustic similarity and listener judgments
With Molly Babel, I have been working to compare various measurements of acoustic similarity used in the imitation and convergence literature to listener judgements of similarity. These measurements include holistic ones, like distance based on MFCC representations, as well as more targeted measures, like distance based on F0 and formants.