Govern with, Not For: Understanding the Stuttering Community’s Preferences and Goals for Speech AI Data Governance in the US and China

Abstract

Current AI datasets are often created without sufficient governance structures to respect the rights and interests of data contributors, raising significant ethical and safety concerns that disengage marginalized communities from contributing their data. Contesting the historical exclusion of marginalized data contributors and the unique vulnerabilities of speech data, this paper presents a disability-centered, community-led approach to AI data governance. More specifically, we examine the stuttering community's preferences and needs around effective stuttered speech data governance for AI purposes. We present empirical insights from interviews with stuttering advocates and surveys with people who stutter in both the U.S. and China. Our findings highlight shared demands for transparency, proactive and continuous communication, and robust privacy and security measures, despite distinct social contexts around stuttering. Our work offers actionable insights for disability-centered AI data governance.