Dr. Josef Fruehwald (University of Kentucky) is the featured speaker for the joint UBC and SFU LinguisticsNOW colloquium over Zoom.
Zoom link: https://sfu.zoom.us/j/66864624518?pwd=bXR1RWNCeW5zc1JmbGxlTGxxMHVlQT09#success
(Meeting ID: 668 6462 4518 | Password: 0928375)
Title: “What should we do when the formant track doesn’t look like the IPA transcription?”
In (socio)phonetics, there are two primary means of collecting and communicating data. The first is close listening and transcription of our percepts. The second is acoustic measurement which is increasingly being automated. The output of such automated acoustic measurement, however, will always be judged successful to the degree it aligns with either the results of close listening, or some other method of acoustic measurement which itself is known to align with the results of close listening. The reason I’ve become concerned with these issues of measurement validity is because recently when I have made dynamic formant plots of /aw/ (Mouth lexical class) drawn from the Philadelphia Neighborhood Corpus using the FAVE suite, the results have not looked like prior descriptions of this diphthong. In this talk, I will walk through my attempts to account for why this might be through a close comparison of automatically extracted formant tracks to formant tracks extracted “by hand” for one speaker. I will explore whether the mismatch is due to systematic error in the FAVE suite, an artefact introduced from statistical methods, or whether, in fact, the formant tracks are accurate, and just don’t align with researcher percepts for some reason.
