Five Years of Gender Representation Analysis in Audiovisual Media: Bridging Research, Policy, and the Public

Abstract

The philosopher Michel Foucault described media as technologies of power that help establish norms while claiming to reflect society. The underrepresentation and stereotypical portrayal of women in the media contribute to their marginalization. Media analysis is thus viewed as a means to objectively expose these representational biases and promote social equality. Quantitative data are often seen as authoritative arguments, essential for raising awareness and driving policy change. However, conducting such studies requires overcoming practical challenges to find acceptable trade-offs between sample size and analytical granularity, especially when working with limited resources. Recent advances in automatic analysis allow for the examination of much larger datasets, mitigating sampling biases associated with manual methods. This exhaustivity, combined with the appeal of machine learning, enhances both the credibility and social impact of the resulting indicators. Several of these automatically produced indicators are now included in official reports that inform public policy—such as MP Calvez’s report on women’s visibility in media during crises, and ARCOM’s (formerly CSA) annual reports on gender representation on TV and radio. This talk reviews the analytical methods developed as part of the ANR-funded Gender Equality Monitor project: automatic estimation of speaking time for women and men [Dou18], visual screen time [Dou24], on-screen text analysis [Dou20], gendered name counts [Dou24], topic modeling by gender [Peil24], and detection of interruptions [Uro24]. Beyond raw numbers, the seminar explores correlations between automatic indicators and manual annotations from sources like the GMMP [Bis24] and ARCOM [Dou24]. It also addresses the technological and theoretical maturity of each analytical method—a key factor in their transfer to public-facing reports with media and societal impact. Finally, it considers future challenges in designing systems that go beyond binary gender representations [Dou23].

  • [Dou18] Doukhan, D., Poels, G., Rezgui, Z., & Carrive, J. (2018). Describing gender equality in french audiovisual streams with a deep learning approach. VIEW Journal of European Television History and Culture, 7(14), 103-122.
  • [Dou20] Doukhan, D., Coulomb-Gully, M., & Méadel, C. (2020). En période de coronavirus, la parole d’autorité dans l’info télé reste largement masculine. La revue des médias, (1).
  • [Dou23] Doukhan, D., Devauchelle, S., Girard-Monneron, L., Chávez Ruz, M., Chaddouk, V., Wagner, I., Rilliard, A. (2023) Voice Passing : a Non-Binary Voice Gender Prediction System for evaluating Transgender voice transition. Proc. INTERSPEECH 2023, 5207-5211, doi: 10.21437/Interspeech.2023-1835
  • [Dou24] Doukhan, D., Dodson, L., Conan, M., Pelloin, V., Clamouse, A., Lepape, M., Van Hille, G., Méadel, C., Coulomb-Gully, M. (2024) Gender Representation in TV and Radio: Automatic Information Extraction methods versus Manual Analyses. Proc. Interspeech 2024, 3060-3064
  • [Pel24] Pelloin, V., Dodson, L., Chapuis, É., Hervé, N., Doukhan, D. (2024) Automatic Classification of News Subjects in Broadcast News: Application to a Gender Bias Representation Analysis. Proc. Interspeech 2024, 3055-3059
  • [Uro24] Uro, R., Tahon, M., Doukhan, D., Laurent, A., Rilliard, A. (2024) Detecting the terminality of speech-turn boundary for spoken interactions in French TV and Radio content. Proc. Interspeech 2024

Bio

David Doukhan has been a researcher at the French National Audiovisual Institute (INA) since 2016. He previously held postdoctoral positions at LIMSI-CNRS and IRCAM, and worked as a research assistant at MIT. An alumnus of EPITA (SCIA class of 2007), he also earned a Master's degree in Acoustics, Signal Processing, and Computer Science Applied to Music from IRCAM.

His research focuses on audio signal analysis and synthesis, machine learning, and digital humanities. From 2020 to 2025, he coordinated the interdisciplinary Gender Equality Monitor project, funded by the French National Research Agency (ANR), which aimed to automatically describe gender representation disparities in French-language media. His work has led to technology transfer initiatives, including contributions to ARCOM’s annual report on the representation of women on television and radio, and to MP Calvez’s report on women’s visibility in media during times of crisis. He is currently exploring the analysis of racist stereotypes in media and the evolution of spoken language since the 1950s.

Additional Information

The seminar will be given in 🇫🇷 French 🇫🇷, and is open to everyone, either in person in the open space of the LRE lab (Paris campus) or online via Teams.