ReDDDoT Phase 1: Planning Grant: Destigmatizing Disfluencies in Speech AI with Grassroots Stuttering Communities

Trained and optimized for fluent speech, speech AI performs poorly for people who stutter (PWS). Even when automatic speech recognition (ASR) systems do manage to transcribe stuttered speech, the resultant transcriptions often remove disfluencies like filler words, inevitably stigmatizing stuttering and denying the option for PWS to have their disfluencies preserved and normalized in transcripts.

This project will partner with StammerTalk – a grassroots community of PWS – to destigmatize disfluencies in speech Artificial Intelligence (AI) by: (a) developing metrics, tools, and techniques to measure, understand, and address fluency biases in existing ASR models, and (b) studying StammerTalk itself as a case study for grassroots AI development that not only produces more equitable and fair AI models but also fosters technical capacity and collectivity within the community. By empowering grassroots, marginalized communities to engage and drive AI initiatives, this project seeks to challenge the existing concentration of AI power by opening up a paradigm for community-led, decentralized AI data collection and development that prioritizes equity, inclusion, and autonomy.

A cross-sector team of academic and community researchers will carry out three strands of activities: 1) technical work to support the StammerTalk community to develop stuttering- friendly speech AI; 2) empirical work to document, analyze, and understand their working model for community-driven, grassroots AI; and 3) co-design work to develop design concepts that integrate more fair and inclusive ASR models in products. Lastly, all three strands will be synthesized to produce a playbook for grassroots AI outlining the steps to community-led data collection, model evaluation, model development, and product co-design as a capacity builder for marginalized, low-resourced communities.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.