Speech variation and speaker profile modeling in media M/F

The proposed topic lies at the crossroads of automatic language processing, corpus linguistics and sociology. It is aimed to answer the following questions: (1) what are the acoustic cues that characterize political discourse? (2) what are the acoustic cues that characterize journalist/interlocutor alignment or divergence (in the broad sense, guest, interviewee, etc.)?

- listening to and selecting subsets of journalistic data relevant to the subject
- implementation of protocols for extracting subsets of data and for acoustic and prosodic analysis in order to highlight the linguistic parameters characterizing the phenomena analyzed
- statistical modeling
- use of speech recognition techniques (forced alignment, pronunciation variants) to explore the data
- modeling parameters from different sources using machine learning techniques
- modelisation according to sociological research questions

This one-year postdoctoral contract is funded by the CNRS as part of an interdisciplinary call. The subject concerns the socio-phonetic and expressive modeling of variation phenomena in multilingual and longitudinal audio journalistic data. For several decades, journalistic speech has provided valuable material for research in speech tecnologies, humanities and social sciences. For example, the former rely on the abundant corpora provided by this type of recording to train speech recognition systems, or to automatically model attitudes and affective states. In the field of social sciences and humanities, phonetics and laboratory phonology use this type of data to model large-scale variation phenomena according to different study dimensions (language, speaker, speaking style, etc.), as does sociology, which uses it to model media bias, for example. In recent years, interdisciplinary studies have emerged that combine automatic processing with scientific questions from linguistics and the social sciences. The post-doctorate topic proposed here is in line with this research direction, and aims to bring together the achievements of the above-mentioned fields in an interdisciplinary framework at the crossroads of automatic processing, corpus linguistics and sociology.