A Gender Action Plan to Make Voice Technology More Inclusive

How Mozilla’s Rebecca Ryakitimbo and the Common Voice team are pushing for ‘respect, equality, and equity’ online

On the surface, voice technology may seem like a female-dominated space. After all, the most recognizable voice assistants — Siri, Alexa — speak with a female-gendered voice.

But these AI voices can be misleading, explains Mozilla Fellow Rebecca Ryakitimbo, a voice technology researcher based in Arusha, Tanzania. “In the voice technology ecosystem, there are fewer women involved than men — especially in Africa,” she explains.

The voices we hear on our smart speakers and phones belie a stark truth: The engineers building these systems are overwhelmingly male. Indeed, when Ryakitimbo was studying Communication Technology and Electronics Technology in university, and later working in the technology industry, she was often the sole woman in the room.

Further, the voice data that trains AI like Siri and Alexa is overwhelmingly male, too. The result? “If you’re having more males working on voice technology, it will represent their needs. It won’t represent the needs of women and others,” Ryakitimbo explains. In practice, this means voice technology has greater difficulty recognizing and responding to female voices. It also means voice technology isn’t built with principles like feminism and gender inclusivity as priorities.

Compounding all this is just how closed the voice technology space can be. “If you look at the majority of big tech companies that work on voice technology, their voice data is not open,” Ryakitimbo says. And so it’s difficult for researchers like herself to truly gauge the severity of the problem — nevermind address it.

The gender disparities that exist within voice technology are especially pronounced in regions like East Africa, where they exist alongside a more expansive digital divide between sexes and genders. In countries like Kenya and Rwanda, women face several obstacles, from less access and literacy than their male counterparts, to poor internet connectivity and prohibitively expensive data bundles. “All these are barriers that are limiting the engagement of women and other gender groups with technology,” Ryakitimbo explains.

Now, as voice technology becomes more and more instrumental to accessing the internet, it’s essential to address this problem. “Voice technology needs to be more balanced,” Ryakitimbo stresses. But how?

Ryakitimbo works on the team at Mozilla Common Voice, an open-source initiative to make voice technology more inclusive. Contributors donate voice data to a public dataset, which anyone can then use to train voice-enabled technology. Ryakitimbo is currently helping to build Common Voice’s Kiswahili data set. “This data set will fuel the first voice technologies to exist in the Kiswahili language,” she explains.

But even in a project as open as Common Voice, disparities persist. “If you look at our English language data, you’ll still find there are more male contributors than female contributors,” Ryakitimbo explains.

To address these problems head on in her Kiswahili work, Ryakitimbo has just debuted a Gender Action Plan — a comprehensive strategy to ensure women and gender diverse communities are equitably represented. The plan will guide gender diversity efforts across Common Voice’s data collection, model creation, and use case development.

(The Gender Action Plan was created by Ryakitimbo after extensive research and community engagement, with help from Mozilla’s Chenai Chair, Britone Mwasaru, Kathleen Siminyu, and Hillary Juma.)

The Gender Action Plan outlines eight challenges around gender within voice systems. Among them: Speech recognition is more accurate for men than it is for women. The majority of developers in the field of voice tech are men. And, issues around privacy and data protection impact how women interact with voice technologies.

In addition to diagnosing the problem, the Gender Action Plan includes a thorough roadmap for how Common Voice — and other voice technology projects — can act. The Gender Action Plan has four key objectives: building a gender-inclusive community of contributors; collecting diverse data; creating diverse models; and ensuring diverse use-case development. Each objective has clear, detailed actions and targets.

Right now, voice technology isn’t as prevalent in East Africa as it is in, say, the United States. And that’s something Ryakitimbo sees as an opportunity for the Gender Action Plan. “There’s a chance to build from the ground up — to cut down on gender bias from the very beginning,” she explains.

The Gender Action Plan will be complemented by a series of events and activities across East Africa. One of them is a “feminist write-a-thon,” Ryakitimbo explains, which will “make sure the sentences and phrases people are contributnig to Common Voice carry feminist messages.” Additionally, Ryakitimbo and the team will launch a gender and voice technology working group, and also run workshops and webinars for women across East Africa. This programming includes courses on how to enter the voice technology space; how to secure funding for a voice technology project; an introduction to Natural Language Processing; and more.

Ryakitimbo has ambitions beyond just reforming voice technology — she also hopes to make progress on gender issues more broadly. The UN’s “16 Days of Activism against Gender-Based Violence” began on November 25, and Ryakitimbo notes there’s a close affinity between this initiative and her own. “I want to make sure actions taking place in the offline world are not happening in the online world,” she explains. Voice technology can be a tool to fight gender based violence, she adds, if it’s built for and by women. For example, it can enable access to crucial information like menstrual hygiene and sex education.

In that vein, the Common Voice team in collaboration with Pollicy will host on December 10 a “16 Days of Activism Digital Resource Share-a-thon.” You can follow this on Twitter with the hashtag #16DaysofActivism.

Voice technology also intersects with a number of other issues, from land rights to agriculture. “People who are less educated can really benefit from voice technology,” Ryakitimbo explains. “If they can’t read or write, they can now access information using their voice.” She gives the example of a farmer in rural Tanzania, who might use voice technology to research fertilizer and crop prices, or to keep records of her harvests.

Although Ryakitimbo’s work is ambitious and far-ranging, she is guided by a simple focus: “Voice technology should be able to recognize women’s voices. It should respect and perpetuate and push for equality and equity for women.”