From Message Composition to Conversation Participation: Bobbai, a Context-Aware AI Speech Application, and Its Implications for Everyday AAC Use

Abstract

Rate of communication remains one of the most persistent barriers in augmentative and alternative communication (AAC). While speaking partners exchange 150-200 words per minute, many aided communicators generate output at a fraction of that speed, limiting spontaneous participation and humor and therefore fun in everyday conversation. Recent advances in large language models and automatic speech recognition are reshaping what is technologically possible for AAC, and raise questions for the field about autonomy, authorship, privacy, and access.

This presentation introduces Bobbai, an AI-supported listening & speech application developed by Lifechaingers (Netherlands) in co-design with people who use AAC, their families, and healthcare professionals, including clinicians involved in ALS care and rehabilitation. Unlike traditional systems that require users to compose messages  letter by letter or word by word, Bobbai listens along with the surrounding conversation, interprets its context, and generates personalized response suggestions aligned with the user’s personality and communication style. The user selects and speaks a suggestion with a single tap, click, or eye movement, delivered through a clear, natural-sounding (personal) voice. The application runs on mainstream devices (iOS, Android, Windows), interoperates with commonly used communication aids, and is currently used by people with ALS, MS, aphasia, cerebral palsy, and acquired brain injury, both at home and in rehabilitation, speech therapy, and daily care settings. User evaluation findings and outcomes from pilots will be shared.

Beyond demonstrating the technology, the presentation uses Bobbai as a lens on the broader implications of conversational AI for the AAC field: (1) the shift from message composition to conducting conversation and participation, and its consequences for conversational timing, repair, and partner dynamics; (2) tensions between AI-generated efficiency and communicative autonomy, authorship, and authentic self-representation, and the design choices that keep the user in control of every spoken word; (3) lessons learned from iterative co-development with users and clinicians, including what changed as a result of user feedback; and (4) open questions concerning privacy of ambient conversation capture, language and cultural coverage, funding pathways, and equitable access.

*This abstract has been reviewed by AI-model Claude Fable 5 for readability and brevity*.