This website offers a visualization of the entire "Speaking time for women and men on television and radio" dataset put online by INA researcher David Douhkan.
The data set was compiled using the free software "inaSpeechSegmenter". The software was developed in collaboration with INA's research department, of which David Douhkan is a member, and the computer science laboratory at the University of Le Mans. The software makes it possible to rapidly analyze a large quantity of sounds, and then to define the gender (H or F) to which the various speakers belong. Finally, it calculates the speaking time distributed between these different genders.
700,000 hours of television and radio programs were analyzed to create this dataset. The analysis was based on audio and audiovisual material from 21 radio stations and 34 television channels, both public and private, covering the period from 1995 to 2019.
The data collected led to the conclusion that, on both radio and television, women speak on average 2 times less than men. In the following article, David Doukhan presents the key data to remember: https://larevuedesmedias.ina.fr/la-radio-et-la-tele-les-femmes-parlent-deux-fois-moins-que-les-hommes
Gender equality, particularly between men and women, is a key social issue. In a society that claims to be egalitarian, we are convinced that it is contradictory for one gender to dominate the production and presentation of information. It then intrinsically brings its own visions and biases.
The way information is received and interpreted today would be very different if the media communicated it on an equal footing.
It's important to point out that this set has been put together according to a binary vision of genre. For this reason, our project is unable to capture the full range of genders and the intersectionality of discrimination present in the media. However, highlighting this aspect of domination is a first step towards understanding the discriminatory our societies.
Today, there are many data sets on the representation of women and men in the media. However, very few of them are as transparent as the one we're working with.
The inaSpeechSegmenter software used to create the dataset is open source (https://github.com/ina-foss/inaSpeechSegmenter). We were also fortunate to be able to talk to David Doukhan, the researcher who designed the dataset. He was able to give us a detailed explanation of the ins and outs and biases of his project.
The major bias of this dataset comes from its training data. Artificial intelligence needs to be trained on annotated data, usually manually. In the case of InaSpeechSegmenter, training was based on the database of speakers referenced in the INA archives. This training database was built up from 32,000 sound extracts broadcast between 1957 and 2012. This corresponds to 1,780 male and 494 female speakers of French.
As a result, this artificial intelligence performs extremely well when analyzing debates or news presentations, media programs for which the AI has an error rate of just 0.6%. However, AI remains very approximate when the media analyzed does not belong to this domain, such as children's cartoons. For this reason, InaSpeechSegmenter only worked on recordings starting at 10 am. To avoid calculating results from cartoons. It's important to note that sung voices have been categorized as music and therefore excluded from the training data. Commercials, on the other hand, have been included.
Generally speaking, artificial intelligence can never achieve perfect results. As a result, the data produced always contain a variable number of errors. It is therefore clumsy to describe them as a perfect representation of reality. Nevertheless, it is possible to extract trends, which can then be analyzed.
Arcom (Autorité de régulation de la communication audiovisuelle et numérique) has also posted data on this subject. Their result is more precise, as it allows results to be obtained according to specific time slots, excluding commercials. However, this dataset was obtained via another AI. Their results are therefore not comparable with those available on our website. In fact, their dataset does not contain the same types and numbers of errors. What's more, Basile Jesset's infrastructure doesn't support or even visualize this data.
The two variable fonts, "WOMAN" and "MAN", were designed to illustrate and visualize the "Temps de parole des hommes et des femmes à la télévision et à la radio" dataset created by David Douhkan using the "inaSpeechSegmenter" open-source software developed at INA.
The drawing of these characters is entirely subjective and oriented to accentuate the interpretation of a reality described by this data set. On average, men have spoken twice as much as women on TV and radio over the last 20 years. This is why the variable font "MAN" is placed above "WOMAN", crushing it. We've also chosen to break away from the usual representations of femininity (soft, round shapes) and masculinity (hard, straight shapes).
We've created an expressive letter design that's closer to an illustration than to a cold, "objective" data visualization (e.g. diagrams, graphs, pie charts). A visualization that could be described as "dehumanized".
The progression of each glyph's design is not linear. The letters metamorphose according to the speaking time they represent. However, we have chosen to give an imposing and overwhelming "MAN" appearance even when the speaking time is less than or equal to that of women. In fact, inaSpeechSegmenter doesn't measure the time at which women speak, or even the importance of the subjects addressed, parameters in which the presence of men also predominates...
The "WOMAN" variable typeface has been designed for economy of growth.
The key features of the different characters are concentrated in the upper part of the glyph.
These characters and their positioning in relation to each other have been designed specifically for this data set.
They would lose their relevance and strength if used in any other context.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Cras sapien ante, sagittis et nunc sed, commodo luctus libero. Vivamus venenatis et justo placerat venenatis. Pellentesque feugiat sem nibh, sit amet venenatis enim finibus a. Sed lacus sapien, congue sed lobortis eu, pretium ut lectus. Vivamus volutpat dolor quis felis pulvinar vestibulum. Praesent vitae enim feugiat dolor consectetur condimentum id et tortor. Etiam nec odio nec nibh aliquet feugiat. Quisque at sodales purus, non feugiat urna. Sed ut felis non felis bibendum fringilla. Integer quis ante suscipit, eleifend massa vitae, eleifend tellus.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Cras sapien ante, sagittis et nunc sed, commodo luctus libero. Vivamus venenatis et justo placerat venenatis. Pellentesque feugiat sem nibh, sit amet venenatis enim finibus a. Sed lacus sapien, congue sed lobortis eu, pretium ut lectus. Vivamus volutpat dolor quis felis pulvinar vestibulum. Praesent vitae enim feugiat dolor consectetur condimentum id et tortor. Etiam nec odio nec nibh aliquet feugiat. Quisque at sodales purus, non feugiat urna. Sed ut felis non felis bibendum fringilla. Integer quis ante suscipit, eleifend massa vitae, eleifend tellus.