Practising with your own photos

Last updated 12 September 2026

Most word-practice apps hand you a set of pictures. A stock mug, a stock dog, a stock bus. This page is about the other way of doing it: photographs of the actual mug, the actual dog, the people whose names they are trying to hold on to, with the word said in a voice they already know.

The short version. In primary progressive aphasia the research consistently finds that practice helps the words you practise, and spreads very little to words you did not. So the list is the whole decision. If the list is generic, the practice is generic. Choose 15 to 30 words the person actually needs, photograph their own things, and record the words in the family's voices.

Why the words have to be theirs

The clearest finding in the PPA treatment literature is also the least convenient one: gains are word-specific. In Henry and colleagues' 2019 study of 18 people with the semantic and logopenic variants, one clinician hour a week plus daily home practice of under 15 minutes produced large gains on the words that were trained, still above baseline at 12 months. Spread to untrained words was small. Doubling the clinician sessions did not add anything; the daily home practice was the part that mattered.

Read that alongside the way most apps are built and the conclusion is uncomfortable. Drilling 500 stock nouns buys you 500 stock nouns. It does not buy you your daughter's name, the word for the thing they need from the top shelf, or the name of the street they live on. Those have to be on the list, by name, or they are not being practised.

So the question stops being "which app has the most content" and becomes "which 20 words would change a day in this house". That is a question only the family can answer.

Why the photo has to be their photo

Two practical reasons, both from the research rather than from taste.

First, transfer. The 2021 semantic-variant consensus review (Suárez-González and colleagues, drawing on 20 years of behavioural research) advises choosing familiar, frequently used concepts and matching the photographs to the objects in the person's own home, so that what is practised on the screen is recognisable on the kitchen counter. A photo of a generic white mug and a photo of the chipped blue one they drink from every morning are not the same card.

Second, engagement. Henry's 2019 protocol used pictures of the participants' own personal objects and credits that autobiographical relevance for people sticking with it. The Communication Bridge-2 randomised trial (Rogalski and colleagues, 2025, 95 people with PPA and their communication partners) went further: the arm that used participant-supplied photos, trained the partner, and set personal goals did better on those personal goals than the arm working from fixed stimuli, 66.7% of goals improved against 49.1%. Both arms did similar amounts of word practice. The difference was whose material it was and who was in the room.

There is also the plain human part. A photograph of a grandchild is a reason to keep going on a bad day. A stock photograph of a child is not.

Why the voice should be someone they know

When the cue finally arrives — the first sound, or the whole word to say along with — someone has to say it. A familiar voice is the better choice, and the research points the same way. Script training studies (the VISTA protocol and its 2024 modified version across all three PPA variants) have people speak in unison with a video of a familiar speaker, a technique called speech entrainment. The home programmes built on look-listen-repeat use a recorded model in the same way.

Practically, a spouse's recorded voice saying "kettle", slowly, is warmer than a synthetic one and easier to speak along with. It also makes the recording session itself a thing the family does together rather than a piece of admin.

Record the same word more than once if you can — a slow version and a normal one. Keep the recordings short: the word, not a sentence about the word.

Choosing the photos

Photograph the real object, in the house, in daylight. Take it from the angle they normally see it from.
One thing in the frame, filling most of it. If you have to explain which part of the picture you mean, it is the wrong picture.
Plain background. A worktop, a table, a wall. Clear the clutter around the object before you shoot rather than trying to crop it out afterwards.
For people, use a photo where that person is alone and their face is clear. A group photo at a wedding is a lovely photo and a poor practice card.
Let the person practising choose as many of them as they can. Involving the family in choosing the material is one of the things a case series of 18 PPA patients named as helping people stay with the programme.
Take more than you need. You will throw some away once you see them full-screen on a tablet.

What makes a bad practice photo

Almost every bad card fails in one of five ways.

Cluttered. A whole kitchen when the word is "kettle". The person has to find the target before they can name it, which is a second task nobody asked for. The supported-conversation guidance is one picture at a time, no competing detail, and the app design rules that follow from it are one photo, one word, nothing else on the screen.
Two possible names. The single most common mistake. Sofa or couch. Pop or soda. Cup or mug. Bag or purse. Coat or jacket. If two people in the family would call it different things, the card is ambiguous and you have built a trick question. Pick the word this household actually says, then make sure the photo shows the thing that word means.
Ambiguous level. A photo of a golden retriever can be "dog", "Milo", or "retriever". Decide which word you are practising, and if it is the dog's name, do not also expect "dog" to count.
Not actually theirs. A picture pulled off the internet of a similar chair. It looks fine and it quietly gives up the reason you were doing this.
Technically poor. Dark, blurry, shot from across the room, or a screenshot of a screenshot. On a tablet held at arm's length these are harder than they look on a phone.

One more, less obvious: avoid photos with writing in them. A packet with the word printed on it turns the card into a reading task instead of a naming task, and you will not be able to tell which one happened.

Picking the first fifteen words

Fifteen is enough to work with and few enough to finish in an evening. The dosage studies used sets of roughly this size: 15 to 30 items in the semantic-variant review, 20 to 40 words in Henry's protocol, cycled in smaller sets of five.

Here is a way to fill the fifteen that usually works.

Five names of people they see or speak to most weeks. Start with the ones they reach for daily.
Four things they use every day and have to ask for: glasses, remote, kettle, keys, stick, phone.
Three foods or drinks they actually want. The thing they ask for at breakfast.
Two places that come up: the name of the town, the café, the church, the ward.
One word they got stuck on this week and minded getting stuck on. There is always one, and it tells you more about what matters than any list you could plan.

Then apply three filters.

Keep about two thirds to words they can still say. Meyer and colleagues (2018) found that gains on words the person could still name at the start persisted up to 15 months, while gains on words already lost had faded by 8. Their advice when time is short is to spend it protecting the words that are still there. It feels counter-intuitive to practise what they can already do. Do it anyway, and treat the lost words as the smaller third.
Short and common first. Word length is the difficulty dial. Long, awkward words early on make the first week harder than it needs to be.
Show the list to the speech therapist. They may want particular words, a particular order of cues, or to swap items out depending on which variant of PPA is involved. Which cues help most differs between the variants, and the therapist is the person who knows which one applies.

Write down, next to each word, the sentence it belongs in and one personal fact about it. "The mug your sister brought back from Spain." That fact becomes a cue later, and it is the kind of thing that is easy to record while you remember it and impossible to reconstruct in a year.

What a session with these cards looks like

The evidence-based shape is short and frequent rather than long and occasional. The PPA studies that worked used something in the range of 10 to 30 minutes, most days of the week, on a small active set. Fifteen minutes daily is the best-evidenced floor. More than half an hour mainly adds fatigue.

Within that, the principle every PPA protocol shares is that the person should not be left stranded in a failed search. The card starts with an unaided attempt, untimed and optional, because retrieval effort is what makes learning stick. If the word does not come, help arrives in a fixed order — a personal memory, the first letter, the first sound, the written word, then the word said aloud to say along with — and the card always ends with them saying it. Nothing is ever marked wrong. The stress research is blunt about why: naming tasks raise measurable stress in people with aphasia, and anxiety is common in PPA, so a practice session that produces failure is a practice session that stops happening.

What Words at Home does with them

This is the part of the tool that is different from what you can buy, so it is worth describing plainly rather than in marketing language.

A household starts empty. There is no stock content. You photograph the words, and the app resizes and stores each photo with the word. You record up to three family voices per word, and you can record the person practising saying it themselves. Words carry a category, a syllable count, a personal memory, and a sentence frame, and each is marked as a word they can still say or one you are trying to bring back. If a word has no recording, the app generates a voice for it once so the card still works, but a recorded family voice is used whenever there is one.

A session runs on a tablet, paired once with a code so the person practising never sees a login screen. One photo, one word, and three large buttons along the bottom for the adult sitting beside them: Help, which brings the next cue, Stop, and Got it. No timers, no scores, no percentages, no red, no counts. The screen looks the same every day and the buttons never move. On days when speaking is too hard there is an easy-day shape and a say-it-with-me mode where every word is spoken in unison with a family voice.

The app decides when each word comes back, using the best cue level from the last session, so the words that are solid appear less often and the shaky ones come round sooner. Everything is logged quietly for the adults: which cue produced each word, unaided attempts, moods, days practised, which words are slipping, and a CSV export to hand to the therapist. None of that is shown to the person practising.

There is an optional speech check that listens to an attempt and answers credit, close, or not sure. It never says wrong, and it is deliberately generous, because automatic recognition of single words from someone with aphasia is unreliable enough that no responsible tool should be scoring pass or fail with it. A coach's tap always overrides it, and it can be left switched off entirely.

An honest note. Words at Home is a practice tool built by a family for a family member with PPA, then opened up so other households could use it. It is not therapy, not a medical device, and not a substitute for a speech-language pathologist. Its design follows published PPA research, but the app itself has never been tested in a trial, so we cannot tell you what it will achieve for anyone. What it does is described above; what it achieves is not something we know. Decisions about care belong with you and your clinician.

This page is general information, not medical advice.