Overlaps and Gender Analysis in the Context of Broadcast Media

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3264–3270 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3264 Overlaps and Gender Analysis in the Context of Broadcast Media Martin Lebourdais, Marie Tahon, Antoine Laurent, Sylvain Meignier, Anthony Larcher LIUM Avenue Olivier Messiaen, 72000 Le Mans, France {martin.lebourdais, marie.tahon, antoine.laurent, sylvain.meignier, anthony.larcher}@univ-lemans.fr Abstract Our main goal is to study the interactions between speakers according to their gender and role in broadcast media. In this paper, we propose an extensive study of gender and overlap annotations in various speech corpora mainly dedicated to diarisation or transcription tasks. We point out the issue of the heterogeneity of the annotation guidelines for both overlapping speech and gender categories. On top of that, we analyse how the speech content (casual speech, meetings, debate, interviews, etc.) impacts the distribution of overlapping speech segments. On a small dataset of 93 recordings from LCP French channel, we intend to characterise the interactions between speakers according to their gender. Finally, we propose a method which aims to highlight active speech areas in terms of interactions between speakers. Such a visualisation tool could improve the efficiency of qualitative studies conducted by researchers in human sciences. Keywords: Speech, Overlap, Gender, Corpus, Medias 1. Introduction The analysis of conversational speech allows char- acterising communication phenomena and specifing the relationships between the involved participants. Such sociological study finds applications in the field of political speech, television debates or journalistic speech (Beattie, 1982). Whether in an interview or a debate, we are facing asymmetric interactions between at least two kinds of participants of different status, such as a journalist or a guest, with complementary roles. In this context, the characterisation of interrup- tions can help to identify gender-related phenomena such as manterrupting (unnecessary interruption of a woman by a man (Bennett, 2015). The Gender Equal- ity Monitoring (GEM) project aims at exploring gen- der representations in French broadcast media. In this framework, one of the main challenges is to bring face to face social sciences representations of gender inter- actions with constrained models suitable for automatic speech processing. For instance, in this work, gender is not considered as a social construct but as a physiolog- ical voice characteristic. The characterisation of interruptions in speech can help to analyse the relationships between speakers accord- ing to their gender and role. When a participant in- terrupts another one, what usually happens when jour- nalists contradict the interviewees (Adda-Decker et al., 2008), it favours the presence of overlapping speech segments (overlaps), defined by the presence of two or more speakers speaking simultaneously. Overlaps structures the conversation and gives clues about how and when the different participants take their speech turn according to their roles. The notion of interrup- tion implies a cognitive action which can not be treated as such by machines. In automatic speech processing, oral communication between several people is usually roughly characterised by a succession of speech turns, where the definition of an interruption …

Content truncated to 500 words