Description automatique du taux d'expression des femmes dans les flux télévisuels français

Abstract

This paper presents an automatic approach, based on speaker gender detection, aimed at describing female speaking time percentage in audiovisual streams. Convolutional Neural Network models (CNN) are used to segment audio streams into music and speech segments, attributed to male and female speakers. Female speaking time percentage is described using 170.000 hours of raw streams corresponding to 21 national TV channels, analysed between 2010 and 2017, and 24 regional TV channels, analyzed in 2016. This description shows that speech time is mainly attributed to male speakers in French TV, and that female speaking time percentage has statistically increased in 8 years for 7 channels.