진학프로석·박사 채용 접수 플랫폼

Context: The Glot-TAL project (ANR-24-CE38-3766) focuses on glottalization in tonal languages of the world. It pursues two main scientific objectives: From a linguistic perspective, the project aims at exploring laryngeal activity in human communication by analyzing the interactions
between pitch control, glottalization, and larynx position in tone production.

This work package (WP1) will specifically focus on tonal languages where glottalization plays a phonological role.

This research examines the cues lying in audio signals, electroglottographic (EGG) data, and larynx tracking to determine the (sub-)phonetic types of glottalized tones—whether produced with a glottal constriction or with a creaky voice—and to identify their differences from non-glottalized tones or other non-phonological creaky voice forms. A thorough understanding of these elements will pave the way for a more refined modeling of the typological characteristics of human glottal behavior.

From a machine learning perspective, in collaboration with colleagues from LLF and LACITO, the project seeks to investigate how deep learning approaches can effectively integrate non-audio signals (specifically electroglottographic EGG) into statistical speech analysis tools. The initial experiments will rely on existing corpora from the Pangloss collection, before expanding to a larger sample of typologically diverse languages (scaling up). This component aims at developing tools for analyzing EGG signals to support linguists in the analysis and documentation of languages, addressing the crucial need for annotation and enrichment of data, especially for tonal languages.