As talkers, we all differ in our pronunciations, resulting in cross-talker differences in the mapping between acoustic cues and linguistic categories and meanings. From previous work, we know that listeners have a remarkable ability to rapidly adapt to the pronunciations of an unfamiliar talker, leading to stable cross-talker perception. What is less known is the specific mechanism(s) underlying this adaptive ability.

A long-standing hypothesis in the literature is that listeners achieve stable cross-talker perception by normalizing the acoustic signal for talker-specific characteristics, related to anatomical differences in talker physiology (e.g., vocal tract length). Numerous accounts of pre-linguistic normalization have been proposed over the years. Widely used in variationist sociolinguistics, sociophonetics, and dialectology, accounts have often been compared and evaluated on how well they reduce category variability in vowel spaces. Less is known about their relative plausibility as models of human speech perception – how well they can explain what humans actually do.

In my thesis, I investigate the predicted consequences of vowel normalization for stable cross-talker perception, using Swedish and English vowels. I approach this question by acoustic analysis, computational models and vowel perception experiments. In addition, I report on the static and dynamic acoustic characteristics of the modern-day Central Swedish vowel space.

Full text and Zoom

If you would like

• the full text from Anna, please email her: anna.persson@su.se.
• to participate on Zoom, please email Catharina Nyström Höög one day in advance: catharina.hoog@su.se.

Anna Persson

Marcin Wlodarczak