Google recently partnered with Howard to launch a project called Elevate Black Voices, which focuses on making automated speech technology (ASR) more functional for Black users of different dialects and accents. The project is the first of its kind as the leaders of the data collection efforts include Howard faculty, instead of consisting primarily of outside researchers.
The main focus of Elevate Black Voices is to create a speech dataset of Black language patterns, dialects and turns of phrases that can be recognized by voice-activated products that we use in our everyday lives like Siri, Alexa, Zoom and Apple Carplay. The team hopes this collaboration will help bridge the gap of racial disparities in technology.
A study from the Proceedings of the National Academy of Sciences (PNAS) found that Black people in the United States often have a poor experience using automatic speech technology compared to white users. This often results in Black people changing their speaking patterns and accents, or code-switching, to be understood by their technology.
Dr. Courtney Heldreth, team leader of Google’s Responsible AI Human-Centered Technology UX (user experience), started the project to understand why people in the Black community struggle to use their authentic voices with automated speech technology.
She sought out a partnership with Howard due to their experience with Black culture, linguistics, and history, and invited academics such as Dr. Gloria Washington, Associate Professor of Computer Science, and Dr. Green, Director of First-Year Writing, to participate.
The project grew from an idea to a team of 13 people, who decided that their target demographic was a national spread of Black voices ranging in age, gender and regional dialects. The team started collecting data in the summer of 2023 by sending out mobile phone surveys that participants could respond to quickly and consistently throughout the day.
According to Washington, the surveys consisted of lighthearted questions like, “How would you refer to this bowl of shrimp and grits?” or “Fill in the blank – I wish a person ____.” Participants could then send a voice recording of their answers and were encouraged to answer as casually and honestly as possible.
The team also utilized a diary study, which followed 30 Black participants for several weeks to learn how they interacted with voice technology. “We found that many black speakers feel othered by voice technology,” Heldreth said. “They feel like it doesn’t belong to them, and they feel like it’s more suitable for folks who are white and speak Standard American English.”
Elevate Black Voices is still a relatively new project, but the team has big plans for the future. Though their first iteration of research ends in December 2023, EBV proposes to continue the project long-term and involve other HBCUs, creating a consortium that can protect the data collected.
“We’re traveling to other HBCUs like Spelman and Morehouse to get people excited about the project, and hopefully they’ll want to sign up and involve themselves in the study,” Washington said. Currently, the project has around 200 participants.
When it comes to defining African American Vernacular English (AAVE), Green said that it’s not just a specific set of vocabulary terms that separates the language from white English.“Black English, like any other language, is a rule-governed system, it includes not just specific phrases or terms, but language patterns. We’ve been looking at regional differences so far, so as you can imagine certain terms, idiomatic expressions and pronunciations will pop up.”
Green used the phrase “I quit school because of recess” as an example. “It’s a phrase that means I don’t play games. We can look at it as a playful expression, but it represents a particular worldview because it’s obvious that it’s a thing that people hold on to when it comes to their daily language,” he said.
Washington referenced the familiar scenario of trying to use Siri while driving in a car. “For example you’re listening to music, trying to figure out your way around, and you’re distracted. I’m from Missouri, my countryness, and my particular dialect is going to come out during that time. I shouldn’t have to say, ‘Where is the closest taco place near me?’ I should be able to say, ‘Yo, can you find me some food?’”
Due to the sensitive nature of data collection in black communities, Heldreth wanted to ensure that the information collected couldn’t be used against them. Howard will retain full ownership of the dataset and licensing to ensure that the data benefits the Black community.
Google is working with Howard to establish a framework for responsible data collection, which determines who can access the data and who cannot.
“When we’re talking about uplifting the voices of the Black community, I think in order to see that through, Black people should own that data,” Heldreth said. “It shouldn’t be owned by a conglomerate or by Google.”
Howard students had positive reactions when it came to the overall impact of the project and the implication of the data on the black community.
“I think it could be helpful because people don’t have to code switch, which can be tiring,” Natane Robinson, a sophomore ceramics major from Boston, said. “It was never something that I considered was needed before. It’s definitely interesting because I feel like it’s so futuristic.”
Khalyla Semexant, a sophomore acting major from Brooklyn, New York, was surprised that the project was the first of its kind. “My first thoughts were, why hasn’t this been done sooner? We all want to be inclusive, but we forgot about that part,” she said. “I definitely think it’s a step forward that they’re thinking about us.”
For prospective participants interested in participating in the study, the Google form link to sign-up can be found here.
Copy edited by Diamond Hamm
