Howard University and Google Launch Landmark Dataset to Advance AI Understanding of Black Voices

Howard University and Google Research have introduced a pioneering dataset aimed at improving how automatic speech recognition (ASR) systems understand and support Black users. As part of Project Elevate Black Voices, this effort seeks to correct longstanding biases in AI by capturing the linguistic richness of African American English (AAE) and its diverse regional variations.

The research team collected more than 600 hours of speech from Black communities across 32 U.S. states, focusing on dialects and speaking patterns often misinterpreted – or ignored – by current voice technologies. These gaps have long forced many Black users to alter their natural speech to be understood by voice assistants.

“African American English has profoundly shaped American culture for centuries,” said Dr. Gloria Washington, co-principal investigator of the project and a researcher at Howard University. “For voice technology to be truly equitable, it must recognize and respect the full spectrum of AAE dialects.”

Current speech datasets frequently underrepresent AAE, largely because users adapt their speech to fit what AI can process. This new dataset provides authentic examples of AAE, giving developers a critical resource to reduce bias and enhance ASR accuracy across diverse communities.

The project marks a significant step toward more inclusive AI – one where voice technology reflects the diversity of the people who use it.