Howard University and Google Research to advance AI understanding of African American English

Howard University and Google Research have partnered to release a groundbreaking dataset aimed at improving how artificial intelligence (AI) systems understand African American English (AAE). a historically underrepresented linguistic form in voice-recognition technology.

Collaboration between Howard University and Google Research

As part of their ongoing initiative, Project Elevate Black Voices, the research team collected over 600 hours of audio from speakers representing AAE dialects across 32 U.S. states.

The project was designed to address the challenges that Black users face when engaging with automatic speech recognition (ASR) systems. The tools that frequently misinterpret or fail to understand culturally specific speech patterns.

Bridging the Speech Recognition Divide

AAE, also known as Black English, Black Talk, or Ebonics, is a rich, culturally grounded linguistic tradition rooted in history, identity, and resilience. Yet many ASR tools struggle to accurately process this dialect due to training on biased or incomplete data.

As a result, Black users often feel pressure to code-switch or alter their natural speech in order to be understood by digital assistants, transcription tools, and voice-enabled applications. This phenomenon diminishes authenticity and underscores systemic inequities in technology development.

Howard and Google Research: Community-Centered Data Collection

To authentically capture AAE speech, the project team hosted curated community events in multiple cities across the country.

These gatherings featured Black panelists—individuals who live and work in the represented communities—leading open discussions on the intersection of Black culture, technology, AI, and innovation.

Following these dialogues, attendees were invited to contribute their voices through a three-week audio data collection effort designed to reflect real-life language use and lived experiences.

Responsible Data Stewardship and Release

The resulting Howard African American English Dataset 1.0 will initially be released exclusively to researchers and institutions affiliated with historically Black colleges and universities (HBCUs).

This phased rollout ensures the dataset is used in ways that prioritize cultural respect, community empowerment, and ethical research practices.

Howard University will retain full ownership and licensing rights to the dataset, serving as a primary steward to ensure responsible usage and alignment with the interests of Black communities.

Google’s Role in Inclusive AI Development

Google, a collaborator on the project, will also use the dataset to enhance the inclusivity of its ASR systems. The tech company routinely trains its voice-recognition models on a wide spectrum of dialects, languages, and accents. This aims to create more equitable AI experiences for all users.

Future access to the dataset by non-HBCU institutions will be reviewed with a focus on researchers whose work aligns with values of inclusivity, equity, and community-driven impact.