An iPad and iPhone app that learns how someone with limited speech says their words, then says them clearly so everyone around them can understand.
It works alongside AAC apps like Proloquo2Go. Everything it hears and learns stays on the device.
Many people with limited speech repeat a word until someone says it back. WordCatch listens for that moment and learns from it, so the family teaches it just by talking the way they already do.
Their word, big and out loud
Tap Playback when they say something you didn’t catch. When WordCatch recognizes a word, it shows it with its picture and says it in a natural voice. When two words sound alike, it asks instead of guessing, and the answer teaches it.
Listening, and finding new words
Turn on listening and WordCatch tells the communicator’s voice apart from everyone else’s, spots the repeat-until-someone-says-it moments, and saves them. When the same sound keeps getting the same answer, it suggests a new word, or adds it by itself after enough consistent answers.
Progress you can show the speech therapist
WordCatch retrains itself overnight while the device charges. A new version is kept only if it scores at least as well on examples a person confirmed; otherwise it rolls back. A one-tap report summarizes every word for a speech-language pathologist or teacher.
Natural voices, chosen by ear
Listen to a sample of each voice, then download only the ones you like. The AI voices run on the iPad and work offline. A family member’s own Personal Voice works too.
AI that runs on the device
Small models on the iPad or iPhone hear when the person has finished talking, tell their voice apart from family and caretakers, and learn a personal model from their own recordings. Natural sentences (“cookie” becomes “I want a cookie”) come from Apple Intelligence, or a small free model you can download. No account, no key, and it works offline.
What it does
Learns from real life. Record words, import audio and videos from Photos, or let it listen and learn when someone says a word back.
Recognizes their words from a few examples each, even words that sound alike, and gets better every night.
Tells voices apart, so it learns only from the person it’s for, and checks with you when it isn’t sure.
Conversation mode: talk as usual while their words appear as sentences and are said out loud, like a live translator.
Playback: tap once, let them talk, and hear what they said. It stops by itself when they finish.
Speaks for them in big type and a natural voice, and asks “ball or bath?” when two words are close.
Brings in their AAC words from screenshots, word lists, board files or PDFs, with pictures from free symbol sets.
Finds new words on its own when the same sound keeps getting the same answer.
Cleans up recordings: move one to the right word, trim it to just the word, or split a run of words into one recording each.
Runs its AI on the device: speech detection, voice recognition, a personal model, and natural sentences, free and offline.
Works across devices: a caretaker records on their own phone and sends the recordings and new words over with AirDrop or Messages.
Shares progress with a speech therapist or teacher as a PDF or spreadsheet.
Works on iPad and iPhone, offline, easy on the battery, with backups to a folder you choose.
It stays on the device
Only their voice is kept
Other people’s speech is turned into text on the device to spot the loop, then thrown away.
Recognition runs on the device
No audio is ever sent anywhere. Sentences are made on the device too; only the optional Claude helper sends a word, never sound.
Backups you control
Scheduled backups to a folder you choose, such as iCloud Drive, with one-tap restore.