research
We carry out empirical work in phonetics and phonology using spontaneous speech corpora. What we build rests on what these studies show, and the instruments the studies require become software in turn.
The relation between speaking rate and F0, the temporal alignment of F0 peaks, and the social distribution of pragmatic intonation markers, tested on large spontaneous speech corpora.
How lexical statistics — neighbourhood density, functional load, phonotactic probability — connect to the phonetic realisation of words, including validation of the measures themselves.
Temporal fluency in learner corpora, discrimination of acoustic features by first-language background, and pronunciation priorities in Korean and English for Japanese speakers.
The analytical infrastructure itself as an object of study: validating forced-alignment accuracy, building measurement pipelines reproducible across corpora, and treating measurement failure as informative missingness.
To keep the papers reproducible we publish aggregate data and analysis code. Source corpora (the Corpus of Spontaneous Japanese, the Buckeye Corpus and others) cannot be redistributed under their licences and are not included; each repository's README states where to obtain them.
Social stratification of pragmatic intonation in spontaneous Japanese: X-JToBI annotations as a sociophonetic resource
Accounts for pragmatic intonation markers — annotated in the X-JToBI labels of the Corpus of Spontaneous Japanese but largely unexploited — in terms of speaker attributes and register. The talk-level aggregate dataset and analysis code are public; the corpus itself cannot be redistributed and is not included.
View the repositorySpeaking rate compresses F0 excursion, not F0 maximum, in spontaneous speech
Shows in Japanese and English spontaneous speech that faster speech compresses F0 excursion by 1.4–1.5 semitones while the F0 maximum stays essentially flat. Reproducible analysis code and manuscript materials are public.
View the repositoryPhonological Neighbourhood Metrics Are Boundary-Sensitive: A Controlled Comparison Framework with Korean as a Case Study
A methodological study comparing five conditions in Korean to show that neighbourhood density and related lexical statistics are not fixed properties of a word but depend on the lexical boundary over which they are computed. The comparison framework and derived data are public.
View the repositoryLexical Importance in L2 Spoken English: Frequency, Phonological Vulnerability, and Spoken Word Recognition
Characterises lexical importance in L2 spoken word recognition through frequency, phonotactic probability, lexical competition risk and phonetic instability. The derived-measure dataset for 2,996 content words is public; the source corpora are not redistributed.
View the repositoryWe take enquiries about studies involving acoustic analysis, corpus construction and annotation design, and the design and implementation of statistical models.
Contact us