How variable is speech, how reliable are single recordings? Assessing the medium-term dynamics of speech through iterative data collection
- Start date: 4 July 2022
- End date: 1 July 2024
- Funder: British Academy/Leverhulme Trust
- Primary investigator: Leendert Plug
This project will kick-start the investigation of medium-term time-dependent intra-speaker variation in speech production: pronunciation variation in the speech of single speakers observed at different times over the course of days, weeks or months (as opposed to hours, or years).
We know this variation exists, but few attempts have been made at quantifying it.
Doing so will provide vital insight into the reliability of speech production studies which rely on single-session recordings: how sure can we be that the timing of that single session does not affect analysis outcomes?
Quantifying the extent of this type of variation requires iterative data collection: recording the same speakers producing the same speech in multiple recording sessions. This project will yield the largest iterative speech corpus to date and offer systematic analysis of time-dependent intra-speaker variation and its relationship to variation from speaker to speaker in relation to vowel pronunciation, speech tempo and rhythm.
This research is highly timely given the current interest in the reproducibility of scientific research and, within the field of phonetics, the ongoing debate about how methodological decisions may affect research outcomes and how representative laboratory-recorded speech is of speakers' habitual speaking patterns [e.g. Wagner et al. 2015 and Roettger et al. 2019, special issues of Journal of Phonetics]. The descriptive statistics on the relationships between intra-speaker and inter-speaker variation on selected phonetic parameters will also be of value to forensic practitioners in relation to forensic voice comparison. The recorded corpus will be made available to other researchers for work on further phonetic parameters beyond the duration of this project.
Publications and outputs
The expected outputs from this project are the corpus -- a digital resource which will be stored in an open-access repository -- and two journal articles reporting on the analysis of selected phonetic parameters.