NSF Convergence Accelerator Track H: An Inclusive, Human-Centered, and Convergent Framework for Transforming Voice AI Accessibility for People Who Stutter

在一个越来越多地通过声控人工智能(Voice AI)导航的社会中，口吃的人-代表着300多万美国人和全球大约8000万人-面临着一个重大障碍：现有的语音AI技术经常无法识别不流畅的语音模式，导致在可访问性、就业和社会包容方面存在一系列不利因素。该项目通过将语音人工智能转变为理解和尊重人类语音多样性的包容性平台来解决这一紧迫问题。通过口吃者的镜头重新想象语音AI，这个项目不仅支持历史上被边缘化群体的事业；它还增强了每个人的语音技术，因为所有发言者都在某种程度上不流利。这些进展与NSF的使命相一致，通过拥抱和理解不同的言语模式，促进科学进步，促进国民福利，并为更具包容性的社会做出贡献。该项目将以人为中心和融合的范式运作，旨在实现四个关键的协同目标：(1)培养多学科和多部门的利益相关者网络，以引导有影响力的结果，(2)阐明以用户为中心的语音人工智能的整体愿景，(3)设计一套全面的自适应语音人工智能解决方案，并为其评估建立试验台，以及(4)起草管理相关语音人工智能风险的指南。利用尖端人工智能技术，该项目将开创包容性培训和测试数据集以及无障碍自动语音识别(ASR)注释的先河，并开发高级ASR深度学习模型。采用的方法包括迭代研究、与最终用户的持续接触以查明现有和未来的障碍，以及对解决方案有效性的严格评估。总的目标是通过一个对所有人都可访问和响应的生态系统来改变语音人工智能，从而确保每个人的声音都能被听到。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

In a society increasingly navigated through voice-activated artificial intelligence (voice AI), people who stutter-representing over 3 million Americans and roughly 80 million individuals globally-face a major barrier: existing voice AI technologies frequently fail to recognize disfluent speech patterns, leading to a cascade of disadvantages in accessibility, employment, and societal inclusion. This project addresses this pressing concern by transforming voice AI into an inclusive platform that comprehends and respects the diversity of human speech. By reimagining voice AI through the lens of those who stutter, this project not only champions the cause of a historically marginalized group; it also enhances voice technology for everyone, because all speakers are disfluent to some extent. Aligning with NSF's mission, these advancements promote scientific progress, foster national welfare, and contribute to a more inclusive society by embracing and understanding diverse speech patterns.Operationalizing a human-centered and convergent paradigm, this project sets out to accomplish four key synergistic objectives: (1) cultivating a multidisciplinary and multi-sectoral network of stakeholders to steer impactful outcomes, (2) articulating a holistic vision for user-centric voice AI, (3) designing a comprehensive set of adaptive voice AI solutions and establishing a testbed for their evaluation, and (4) drafting guidelines for managing associated voice AI risks. Harnessing cutting-edge AI technology, the project will pioneer inclusive training and test datasets as well as annotation for accessible automatic speech recognition (ASR) and develop advanced ASR deep learning models. The adopted methodology encompasses iterative research, continuous engagement with end-users to pinpoint existing and future barriers, and stringent evaluations of solution efficacy. The overarching goal is to transform voice AI, through an ecosystem that is accessible and responsive to all, thereby ensuring that every voice is heard.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.