AI speech and language rehabilitation

What if recovery could know your name?

Aphonos builds conversational software for people who lost their language to a stroke. A consumer companion today, on a defined path toward a prescription digital therapeutic for post-stroke aphasia.

áphonos · Greek · voiceless

2M+
People living with aphasia in the US — the most common language disorder you've never heard of
~3 hrs
Average weekly therapy time for aphasia patients — a fraction of what research suggests is needed
0
FDA-authorised prescription digital therapeutics for aphasia. There is no device classification, and no product code

The problem

A massive unmet need, hiding in plain sight

Aphasia is the loss of language following a stroke or brain injury. It affects more people than Parkinson's disease, cerebral palsy, or muscular dystrophy. Most people have never heard of it.

Existing tools either digitise traditional drills or help patients compensate for lost language. None of them is built around who the patient actually is, and none has been taken through the FDA as a treatment for aphasia.

That is the gap Aphonos is building into.

180,000 New aphasia cases in the US each year, primarily from stroke
No predicate FDA has never classified a device for aphasia treatment. A search of the 510(k) and De Novo databases returns no aphasia product code and no classification regulation, which means the category has to be opened by a De Novo
Caregiver burden Families shoulder most of the recovery work with no clinical training and almost no support infrastructure

Our approach

Three things no one else is doing together

Aphonos combines conversational AI, pre-stroke identity personalisation, and a caregiver layer into a single platform, built from the ground up for this patient population and grounded in established aphasia treatment approaches.

I

Identity-first personalisation

Every session is shaped by who the patient was before their stroke: their career, family, hobbies, and life. This is not a feature. It is the architecture.

II

Conversation, not drills

Practice adapts in real time to the person's severity, fatigue and progress, structured around Semantic Feature Analysis, constraint-induced language therapy and script training.

III

Caregivers as participants

After every session, families receive a clear summary and practical guidance. Recovery is a team effort. We built the platform that way from day one.

Pipeline

One platform, two regulatory worlds

A consumer wellness product is live today. The clinical programs behind it are headed for a different bar entirely. The dashed line is where that changes, and nothing to the right of it is authorised yet.

Program Indication and intended use Route
Research &
discovery
Development &
feasibility
Clinical validation
pivotal RCT
Regulatory
submission
Authorised /
listed
Reimbursed &
available
APH‑101 Aphasia.ai companion Chronic post-stroke aphasia. Daily conversational language practice at home. General wellness. No clinical claim. General wellness Not a medical device
Live with pilot users and clinician partners
APH‑102 Clinician companion Between-session practice data for the treating speech-language pathologist. General wellness. No clinical claim. General wellness Not a medical device
In development
APH‑201 Unnamed Post-stroke aphasia, naming and connected speech. Intended as an adjunct to clinician-supervised speech-language therapy. US · De Novo Anticipated. No predicate device exists
Feasibility study in design
NIDCD SBIR Phase I
APH‑301 Unnamed Post-stroke aphasia. German statutory reimbursement market. DE · DiGA, CE IIa Planned
Research
Left of the line:  general wellness, not a medical device Right of the line:  regulated medical device, none authorised

Regulatory pathway and device classification are subject to FDA feedback, and any study qualifying as pivotal is subject to FDA feedback and review of data. Programs in research and development are not FDA-cleared, not FDA-authorised, and are not available for prescription. APH‑101 and APH‑102 are general wellness products. They are not intended to diagnose, treat, cure, mitigate or prevent any disease.

The consumer product

Aphasia.ai

A conversational companion for stroke survivors and their families, available directly with no clinical referral. It is where people practise talking on the days they are not in a clinic.

It also does something strategic. Real conversational interaction at home generates the engagement and usability evidence that a feasibility study needs, which is what de-risks the clinical program behind it.

Visit Aphasia.ai →
Aphasia.ai · session preview
Aphasia.ai
Maria, you mentioned you used to teach at Lincoln High. Tell me — what was your favourite subject to teach?
Maria
Math… I… loved… algebra.
Aphasia.ai
Algebra — that's wonderful. What did you love most about it? Was it the puzzles, or watching your students figure it out?

Evidence and collaborators

What we can point to today

An early company's credibility is the checkable part. Named collaborators, named methods, and a regulatory position we can defend, rather than user counts.

Clinical collaboration

A feasibility study is planned with an academic aphasia centre as the study site, with a named speech-language pathologist as collaborator. We will name the site and the investigator here once the collaboration is announced.

Application submitted to the National Institute on Deafness and Other Communication Disorders, September 2026.

Grounded in established methods

  • Semantic Feature Analysis
  • Constraint-Induced Language Therapy
  • Script training
  • Consistent with the Life Participation Approach to Aphasia

Regulatory position

There is no FDA-cleared predicate device for aphasia treatment. The existing digital therapeutic classifications are indication-locked, including 21 CFR 882.5801, which is limited on its face to psychiatric conditions.

The anticipated route for APH‑201 is therefore a De Novo classification request, which would open the category and set the special controls every subsequent product has to meet.

Built on real infrastructure

Streaming speech recognition tuned for disordered speech, a two-layer language model architecture, and a session record designed for clinical review from the start.

Speech recognition is never the only input. Every exercise has a parallel path that does not depend on the system understanding what was said.

Get in touch

Let's talk about what we're building.

Whether you're an investor, a speech-language pathologist, a clinical researcher, or exploring a partnership, we'd like to hear from you.

Please add your name, a valid email, and a message.

✓ Message sent. We'll be in touch shortly.