Now in beta · Free for early testers

Speak the way you can.
Be heard.

AphaSay listens to broken speech from aphasia and reads back the sentence the person was trying to say. Real time. In your own voice.

Join the Beta — Free See how it works →

FAST Check stroke triage

AI speech reconstruction

Selfi face-tracking exercises

Quick Phrases library

Works fully offline

2M+

people in the US living with aphasia

180K

new cases diagnosed every year

80%

AI sentence reconstruction accuracy

The Problem

After a stroke, the words don't come out right.

Aphasia damages the part of the brain that turns thoughts into language. The mind is still sharp. The thoughts are still there. The words just come out wrong, broken, or not at all.

A person with aphasia might say "fork" when they mean "spoon." They might say a non-word like "blurk" when they mean "phone." They might know exactly what they want but only manage three fragments before they give up.

Their family stops asking. Friends visit less. The person stops trying.

Live reconstruction

INPUT“want blurk kitchen wife cold”

PROFILEwife: Fatima, fridge, cold water 3pm

CONFIDENCE94%

OUTPUT"I want my wife to bring me something cold from the kitchen."

Playing reconstruction…

Features

Everything you need.
Nothing you don't.

🚦

FAST Check — Stroke Triage

3-step guided assessment (Face droop, Arm drift, Speech slurring) with live MediaPipe face-tracking and one-Euro smoothing. Practice mode for caregivers, disclaimer gating, and triage history on the Dashboard.

🧠

AI Speech Reconstruction

Whisper transcription → context-aware GPT-4o reconstruction → confidence tiering. Inline 'Did you mean…' alternatives, each tappable to speak. 80% accuracy on aphasic utterances in 2025 peer-reviewed research.

🔊

Speak-This Playback Chain

Reconstructed sentences play back in your own voice. Pipeline: ElevenLabs cloned voice → OpenAI TTS → on-device expo-speech fallback. Voice cloning from recorded samples — no specialist hardware required.

🙂

Selfi — Face-Tracking Exercises

Calibration captures your baseline jaw, lip, and symmetry geometry. Four exercises: Open Vowel (AH), Spread Vowel (EE), Round Vowel (OO), Smile Hold. Live overlay, rep counting, symmetry score per session.

📖

Read Aloud — Read Anything

Paste text, open a PDF, scan paper, or import an email — the app reads it back word-by-word, karaoke-style, at a calm pace with pauses at commas. Tap any word to replay, loop a sentence, or get plain-language simplification. Built for reading fatigue, zero-pressure and one-tap.

🗣️

Speak It For Me

Type or tap a phrase and the phone says it aloud for you — at the doctor, the counter, or on a call. Emotional tone selection (calm, confident, friendly, professional), discreet earbud playback, auto-repeat, and quick-access emergency phrases. Speaks in your cloned voice.

💬

Quick Phrases Library

Pre-built phrases across 8 categories: greetings, basic needs, medical, emotions, daily, social, emergency, and custom. Fully editable. One-tap playback in your cloned voice — even offline.

🏠

Dashboard & Progress

Time-of-day greeting, daily conversation goal + progress bar, FAST status chip, stat tiles (total sessions, 7-day avg confidence, Selfi count), AI Coach tip card, and recent activity feed.

👤

Patient Profile & Context

Aphasia type, severity, daily routine, family members, home layout, and communication preferences power the reconstruction engine. The more context the AI has, the more accurate it gets.

🔒

Privacy & Security

All face processing is on-device. Supabase auth, MMKV-backed local stores, SecureStore for API keys. Audio encrypted at rest and in transit. No ads, no data sales, and we never train AI models on your conversations.

Problem guides

Problems we help with.

Plain-language guides for speech, language, reading, writing, and everyday communication after stroke.

See all communication problems

Aphasia after stroke

A plain-language guide to aphasia after stroke, why speech can break down, and how families can support communication.

Read guide

Dysarthria after stroke

How weak or uncoordinated speech muscles affect clarity after stroke, plus practical intelligibility strategies.

Read guide

Apraxia of speech after stroke

A guide to motor planning problems that make speech effortful, inconsistent, or hard to start after stroke.

Read guide

Word-finding problems after stroke

Why tip-of-the-tongue moments happen after stroke and how cueing, context, and backup tools can help.

Read guide

Understanding speech after stroke

What receptive aphasia and listening fatigue can look like after stroke, and how to make conversation easier.

Read guide

Reading and writing after stroke

How alexia, agraphia, fatigue, and aphasia can affect reading, writing, texting, and paperwork after stroke.

Read guide

How It Works

Speak however you can.
We figure out the rest.

Listen

Tap the microphone and speak. The app records whatever comes out, even if it's broken or full of wrong words.

Reconstruct

The AI takes what you said and combines it with what it knows about you, your family, and your daily life to figure out what you meant.

Speak

The app reads the reconstructed sentence aloud in a clear voice. With voice cloning, that voice can be your own.

Built on GPT-4o, which scored 80% accuracy on aphasia speech reconstruction in a 2025 peer-reviewed study in Scientific Reports.

See the full AI pipeline →

See it in action

Watch the overview.

A quick look at how AphaSay works — from opening the app to hearing your words spoken back in your own voice.

▸Tap to speak — even if the words come out broken

▸AI reconstructs the sentence you meant in real time

▸Plays it back in a clear voice you can clone to your own

More feature videos on the Tutorials page →

The App

Designed to be used
with one hand.

FAST Check — stroke triage

AphaSay Dashboard — daily goals and stats

Dashboard — goals, stats, AI coach

Speak — AI reconstructs your words

AphaSay Selfie — face-tracking exercises

Selfie — live face-tracking exercises

Want to try it right now? Join the beta waitlist for instant access on Android today — iPhone coming soon.

Examples

What AphaSay actually
sounds like.

Asking for help

Patient said

"phone... wife... call... now"

AphaSay said

"Please call my wife on the phone now."

Daily request

Patient said

"want blurk... cold... kitchen... thing"

AphaSay said

"I want something cold from the kitchen."

Medical

Patient said

"head... hurt... medicine... need"

AphaSay said

"My head hurts. I need my medicine."

Evidence Base

Built on clinical research.
Not guesswork.

AphaSay's exercise content, scoring methods, and feedback system are grounded in published speech-language pathology research.

🧬

Neuroplasticity & frequency

Neuroplastic recovery responds to practice frequency over intensity — supporting our daily short-session model over weekly marathon sessions.

Bhogal et al., Stroke, 2003

🎙️

Motor speech feedback loops

Real-time visual and auditory feedback during speech exercises improves motor learning retention. AphaSay provides both — camera overlay and audio scoring — on every rep.

Maas et al., American Journal of Speech-Language Pathology, 2008

📐

Facial rehabilitation & symmetry

Structured physical rehabilitation improves facial symmetry and function after facial palsy. AphaSay tracks symmetry automatically across every session.

Vaughan et al., Frontiers in Neurology, 2020

🤖

AI reconstruction of aphasic speech

La Trobe University's 2025 study showed GPT-4o correctly reconstructed 80% of aphasic utterances from AphasiaBank data — the research foundation for AphaSay's Talk mode.

Adikari et al., Scientific Reports 15:24725, 2025

AphaSay is not a medical device and does not replace professional speech-language pathology services. Always consult a qualified SLP for clinical diagnosis and treatment planning.

FAQ

Common
questions.

Can't find your answer? Email us and a real person will reply within 48 hours.

Privacy

Your voice. Your data.
Nobody else's.

Audio is encrypted on your device and on our servers. Recordings are deleted on a schedule you control. We do not sell data. We do not show ads. We do not train AI models on your conversations.

On-device face processingEncrypted at rest & in transitNo ads. No data sales.Never used to train AI models

Read the Privacy Policy →

Start speaking again.

Beta access is free and open now. Join in under 30 seconds.

Join the Beta — Free

Other apps for stroke recovery and independence

Tools designed to make recovery more doable every day.

Speak the way you can.
Be heard.

After a stroke, the words don't come out right.