AAC Awareness Month Roundup: The AI Tools Pushing Past the Legacy AAC Ceiling

Every October, AAC Awareness Month asks the community to look at how far communication technology has come — and anyone who's spent time with a legacy symbol board or a static word-prediction keyboard knows the honest answer is: not far enough, fast enough. Grid navigation is slow. Vocabulary sets are generic. And building a personalized board has traditionally meant hours of a speech therapist's or caregiver's unpaid time.

This year's roundup looks at three projects — a research prototype, a youth-founded nonprofit app, and a commercial AAC platform — that are each attacking a different piece of that legacy bottleneck with AI. None of them are finished products solving every problem. But together they sketch a genuinely different ceiling for what AAC can do.

R-Speak: AI that interprets aphasic speech itself, not just text

Most AI-in-AAC work assumes the user is typing. R-Speak, a co-designed prototype out of Sheffield Hallam University, starts from a harder problem: people with expressive aphasia often can speak, but their words come out disordered, incomplete, or substituted — and current tools don't know what to do with that.

R-Speak uses large language models to interpret fragmented or aphasic speech itself and reconstruct it into coherent, intended sentences — a meaningfully different job than predicting the next word in a clean input. The team tested eight LLMs against real transcripts from AphasiaBank before landing on Mixtral (8x7B) for their first working prototype, which they then tested with four people with aphasia and one care partner, plus a six-person clinician focus group.

The results are a genuinely hopeful signal: participants with aphasia rated the tool as good on the System Usability Scale (mean score of 75) and expressed real enthusiasm about what it could mean for their independence — more so, notably, than the clinicians in the study, who were more cautious about real-world benefit. That gap is worth sitting with: it's the people living with aphasia who saw the clearest upside.

The research team didn't stop at "does this work" — they went on to test 12 lightweight, open-weight models (as small as 0.5B parameters) to find versions fast and light enough to run practically, eventually finding that a 3-billion-parameter model (Qwen 2.5:3B) hit a strong balance of accuracy and near-instant response time. That's a meaningful detail: it suggests this kind of support doesn't require massive cloud infrastructure to be clinically useful, which matters enormously for cost and access. Next up for the team: refining the prototype into a full phone app and testing it with a larger, broader group of people with mild-to-moderate aphasia.

ConnectAAC: built by a teenager who saw the gap firsthand

If R-Speak represents academic rigor, ConnectAAC represents something else entirely: lived urgency, built fast. ConnectAAC was created by Ayana Singh, a 16-year-old founder who started the youth-led nonprofit Equal Speaks after watching a family member run into exactly the communication barriers this whole genre of app is trying to solve.

ConnectAAC combines visual communication cards, personalized user profiles, text-to-speech, and — the part that separates it from a standard symbol board — context-relevant word and sentence suggestions. Instead of making a user hunt through nested menus to find the word they need, the app tries to surface likely language based on the situation at hand, across settings like home, school, healthcare, and the community. It's available as a web app and on Android, free to the families and clinicians who need it.

What makes this one worth including isn't just the technology — plenty of apps offer predictive suggestions — it's who built it and why. ConnectAAC is explicit that it's a supplementary resource, not a replacement for individualized clinical evaluation or an established AAC system, which is the right kind of humility for an emerging tool. But as a demonstration that the barrier to building genuinely helpful AAC tech has dropped low enough for a teenager with a cause to ship something real, it's a meaningful marker of where this space is headed.

Spoken: AI word prediction at scale, with 300,000 users and counting

Spoken takes a more mature, commercial approach — and its scale is the point. The app uses AI-powered word prediction paired with natural-sounding text-to-speech, and it's been used by more than 300,000 people navigating aphasia, autism, cerebral palsy, ALS, Parkinson's, stroke, and other speech and language conditions.

The core pitch is personalization over time: Spoken learns a user's individual speech patterns and adapts its predictions the more it's used, rather than offering one static vocabulary set to everyone. Users can also actively teach it about themselves to sharpen its suggestions — echoing exactly the kind of personalized-context feature that researchers in this space (including the CHI '23 "speech macros" study we covered last time) consistently find AAC users want most.

Spoken's user reviews tell a consistent story: people describe it as the difference between being stuck mid-sentence in a hospital bed and actually being able to tell a family member they love them. That's the unglamorous, high-stakes use case that legacy AAC tools have struggled with — not because the hardware couldn't do it, but because static, generic vocabulary sets simply aren't fast or personal enough in the moment that matters.

The throughline

What connects all of these — from a university research prototype to a teen-founded nonprofit app to a 300,000-user commercial platform — is that each is chipping away at a different piece of what made legacy AAC so exhausting: rigid vocabulary, slow navigation, and one-size-fits-all design. None of them have solved every problem AAC users have raised about AI (privacy, hallucination, and loss of personal voice remain live concerns across this research, our own past coverage included). But the direction is unmistakable: AAC is moving from static and generic toward adaptive and personal, and that shift is worth genuinely celebrating this Awareness Month.