Howard University and Google Research have launched Project Elevate Black Voices, a new partnership aimed at making voice technology more inclusive by training artificial intelligence to better understand African American English (AAE). The project addresses long-standing issues Black users face with automatic speech recognition (ASR), which often misinterprets or fails to recognize natural speech patterns in Black communities.
The team collected over 600 hours of speech data from Black speakers across 32 states, creating a high-quality dataset that reflects the richness and diversity of AAE. Researchers found that many Black users “code-switch” to standard English to be understood by voice assistants like Siri or Alexa, which leads to inaccurate, inauthentic experiences with the technology.
“It’s about time that we provide the best experience for all users of these technologies,” said Dr. Gloria Washington, Howard University computer science professor and co-principal investigator. “Voice assistant technology should understand different dialects of African American English.”
The lack of authentic AAE in existing datasets is a key factor in the racial bias built into ASR systems. Even when data is available, code-switching and privacy policies often make it unusable. To address this, researchers hosted culturally grounded events in Black communities, where participants contributed their voices after open discussions on A.I., identity, and technology. These efforts resulted in organic, community-driven data collection.
“We want to make sure we aren’t getting left behind in the improvement of technology,” said Dr. Lucretia Williams, project lead and senior research scientist at Howard.
Howard University will retain full ownership and licensing rights to the dataset, which will first be made available to researchers at HBCUs. Google will also use the data to improve its voice recognition tools. Any future broader release will prioritize organizations aligned with values of inclusivity, empowerment, and community-focused research.
African American English—also known as Black English or Ebonics—is a cultural language rooted in Southern traditions, widely spoken across the U.S. Despite its influence on American culture, AAE has often been misrepresented or ignored in tech development. This project aims to change that, ensuring that Black voices are accurately understood and authentically represented in the digital world.
