Practical Corpus Phonetics

Presenters: Michael McAuliffe, Morgan Sonderegger

T1TFD3

Corpus phonetics, the study of speech production in non-laboratory settings, has become a major approach in phonetics and phonology research, in contexts from fieldwork data with small numbers of speakers to large-scale cross-linguistic studies of thousands of speakers. This workshop aims to bridge the gap between the availability of corpus phonetic tools and their practical application. We will survey computational tools for constructing and working with speech corpora to answer linguistic questions, including automatic transcription (e.g. Whisper), speaker diarization (PyAnnote), text-to-speech alignment ("forced alignment": Montreal Forced Aligner), speech database systems for representing and querying corpora (PolyglotDB), and automatic phonetic measurement (e.g. for vowel formants, VOT). The workshop will be centered on hands-on labs where participants gain experience with this rapidly-growing ecosystem of tools. We will discuss case studies and best practices for large-scale corpus studies. Participants are welcome to bring their own data for the labs or use data provided by the presenters.

Keywords: Phonetics, Phonology, Corpus Linguistics, Communicative Efficiency, Quantitative Methods, Computational Linguistics

Presenters

Michael McAuliffe

McGill University

My primary research interests center on creating tools for leveraging large amounts of speech data to answer phonetic questions. I currently maintain the Montreal Forced Aligner (MFA), a tool for aligning speech to text, and most of my current research involves improved alignment models and techniques. I also was the primary developer for PolyglotDB and ISCAN, tools for acoustic analysis of large speech corpora as part of the SPeech Across Dialects of English (SPADE) project. My other research interests are in speech perception, perceptual learning, and quantifying acoustic similarity.

Morgan Sonderegger

McGill University

Morgan Sonderegger is Associate Professor of Linguistics at McGill University and Canada Research Chair in Speech Variability. His research focuses on phonetics, sound change, and phonology, primarily using corpus data. Another focus is methodology for quantitative linguistic data: both statistical analysis (e.g. "Regression Modeling for Linguistic Data": MIT Press, 2023) and software for speech database management and automated speech analysis (Montreal Forced Aligner, PolyglotDB, AutoVOT).