Short Description

This thesis aims at investigating the acoustic-prosodic alignment in conversational speech. More specifically it studies adjacent turn pairs of two speakers and relates the acoustic-prosodic characteristics of these turn pairs to lexical and semanitc similarity measures. The hypothesis is, that in cases of high semantic similarity, and even more so for cases of high lexical similarity, speakers will also show a matching prosodic configuration, and potentially also match with respect to pronunciation variation. While most previous research studied global alignment over whole conversations between strangers, the focus of this thesis is on alignment between friends, partners and colleagues as a more local phenomenon. Finally, this thesis is a quantitative, more general continuation of our earlier studies, in which we showed that (dis)agreement and preference Ssructure are reflected in alignment along distinct acoustic-prosodic features (10.21437/Interspeech.2023-1538).

Your Tasks:

• Extract acoustic, lexical and semantic features from the Grass Speech Database
• Train different types of classifiers suitable for small data sets (e.g., Random Forests)
• Use Shapely tools to investigate the features’ importance
• Analysis of the classification and documentation of your results (thesis writing)

Your Profile

• basic knowlegde of sound engineering and/or speech communication
• good knowledge of programming (e.g., Python)

Contact:

Barbara Schuppler (b.schuppler@tugraz.at)