# Reading Adventure — proposal for a first app

## Implementation update — 8 September 2026

An English pitch and fully English demo are now implemented. The practice app adds four candidate sessions per language, parent-selected activities and a factual session recap. Saving the last 50 session records and parent choices in the current browser is optional and off by default. English and Dutch records are separate. Speech recognition, automatic reading assessment, expert approval and demonstrated learning gains remain future work. The roadmap below describes the broader product ambition; features beyond this update are proposals.


September 8, 2026 · Working name, product decisions and a testable development plan

## Value proposition

**Reading Adventure helps children practise short reading activities at home, with support when they get stuck and a clear recap for their parent.** Start with Dutch-speaking children aged approximately 6–9 in Flanders who are already receiving reading instruction. A parent helps them get started; practical testing must establish how independently children can use the app.

The first version aims to make suitable practice more accessible. Learning gains, pricing and independent assessment of reading aloud still need to be investigated. We intend to develop the product with input from education and speech and language therapy specialists; there are no confirmed partners yet.

This English website and its demo are a presentation adaptation. The initial educational target remains Dutch-speaking children in Flanders. The English samples introduce the concept; they are not a validated English reading programme. Educational validation of the original Dutch learning content is also still pending.

## Purpose and initial audience

The app supports practice at home alongside school and any speech and language therapy the child receives. Activities build on letters and sounds the child has already learned. The target group, starting age, reading confidence and independence are working hypotheses to test.

Reading and pronunciation follow separate development tracks. The first version focuses on letter–sound correspondence, reading words, short sentences and comprehension. Specific articulation exercises will follow only with specialist guidance and tested speech analysis. A transcript containing the intended word does not prove that every sound was pronounced correctly.

## What the child does

As a design starting point, a short session lasts five to ten minutes. This is a usage choice to test, not a prescribed treatment duration. The child can pause or stop at any time.

1. **Get started:** familiar buttons and brief instructions; the parent selects suitable material.
2. **Review:** familiar letters, sounds or words return.
3. **Practise:** words and short sentences, with one activity per screen.
4. **Get help:** listen, read together or try again. An answer given after help does not count as independently mastered.
5. **Understand:** one or two questions about what the child has read.
6. **Finish:** a factual recap recognises the effort of trying. There are no leaderboards or penalties for missing a day.

English presentation example: a child who knows the necessary letters and words reads “The fox is in the den.” Help is available for a word such as “den”. The app then asks: “Where is the fox?” Future sessions would also use different words or a new sentence, so the programme does not simply reward memorised answers.

## Scope of the first version

Distinguish the clickable concept on the pitch website from a pilot that families can use at home.

| Component | Clickable concept | MVP for a supported pilot |
|---|---|---|
| Start | One fixed sample session with the mascot Fox and a woodland theme; a parent briefly reads along. | The parent chooses familiar letters and words from a limited, reviewed selection. |
| Learning material | Two words (“fox”, “den”), “The fox is in the den.” and one comprehension question: “Where is the fox?” | Approximately ten reviewed sessions, including review and new texts. The exact amount depends on content review. |
| Help | Listening if a suitable device voice is available, help text, skipping and pausing. | Reviewed audio examples and fixed hints for each word, available on the chosen test devices. |
| Responses | “I gave it a try” records an action; only the fixed comprehension question receives feedback on the answer. | Answer choices and parent observations remain clearly distinguishable from any speech observations. |
| Recap | Shows what was clicked, attempted and skipped. | The parent sees practice activity, help and review; simple local progress that can be deleted is sufficient for the first pilot. |
| AI | Describes intended capabilities; does not simulate speech analysis. | Initially a tool for preparing reviewed content. Automated reading feedback follows only after a separate, successful speech evaluation. |

The MVP does not include open chat, diagnosis, articulation therapy, public leaderboards, payments or automatic school integrations. A web app for tablets and computers is the starting point. A native mobile app and an extensive teacher portal are later additions.

The sample sentences are not a reading-level assessment. Even short words such as “the” and “is” require appropriate prior knowledge. A reading specialist must therefore design the actual learning sequence; an age label or short word length is not enough. The English demo is adapted for presentation and does not establish that either language version is educationally validated.

## The role of AI

AI can help prepare reviewed story variations, select suitable activities and later support reading aloud. Initial activity selection can use simple rules. Learning material, hints and audio examples are reviewed before children use them; recordings of individual sounds require specialist review.

Three concerns shape the design: recognition errors, unsuitable AI explanations and unnecessary data storage. Use a bounded activity bank, an explicit “uncertain” outcome and minimal data processing. Unclear audio should produce a response such as: “I couldn’t hear you clearly.” This does not lower the child’s reading level. A Flemish accent or another home language is not automatically an error.

The speech evaluation must include sounding out words, hesitation, self-correction, repetition and background speech. A recogniser must not silently replace a reading error with the expected word. A specialist determines which variants and error categories matter. Transcription and reading-error detection need separate evaluation; research with children in the Netherlands does not establish reliability for the Flemish target group. See the feasibility note for technical detail.

## What the parent sees

- Simple settings for suitable letters, practice topics and audio.
- An overview of material practised, first attempts, help used and review.
- A distinction between answers observed by the app, uncertain recognition and the parent’s notes.
- A short recap that the parent can choose to share with the school or a speech and language therapist.

Starting entirely independently is a later usability milestone. Begin practical testing with an adult nearby and measure how much help is still needed. Persistent difficulties may prompt a conversation with the school or a speech and language therapist; the app does not make diagnoses.

## Development sequence

**Step 1 — a usable practice prototype.** Build one complete session and a factual parent overview. Identify which feedback comes from fixed answer choices. The prototype works without cloud storage or voice recording and tests the practice experience.

**Step 2 — a separate speech evaluation.** Compare a candidate speech recognition system with human assessments, with consent and clearly bounded processing. Set acceptance criteria in advance. Only a successful evaluation justifies automated feedback for children.

**Step 3 — a small supported pilot.** Proposal: two speech and language therapists, two teachers and ten to twenty families. These intended participants are not confirmed. The pilot tests usability and does not, by itself, demonstrate effectiveness.

**Step 4 — investigate learning gains and expand.** Investigate transfer to new reading activities outside the app. Then add further learning material. Articulation remains a separate development track.

## Success criteria and decision points

The criteria below are proposed decision rules, not achieved results. Set them before testing and report numbers of children alongside percentages in a small pilot.

| Question | Measurement | Initial decision rule |
|---|---|---|
| Is the content suitable? | A teacher and a speech and language therapist review all activities, hints and audio. | All available content has been reviewed; unresolved content concerns prevent the affected activity from being used. |
| Can a child use the interface? | Observe starting, finding help, pausing and finishing after a brief shared introduction. | At least 8 of the first 10 children finish a session with no more than one additional instruction about operating the interface. Help with learning content is counted separately. |
| Is the practice experience comfortable? | Ask the child for a simple reaction afterwards and observe stopping or frustration. | Address repeated difficulty or frustration first; completion figures alone do not justify proceeding. |
| Does the parent understand the overview? | Ask the parent to explain what “attempted”, “with help” and a comprehension response mean. | At least 8 out of 10 distinguish practice activity from demonstrated reading mastery. |
| Is speech feedback reliable enough? | Compare with independent human assessments; break down reading errors, incorrect corrections and uncertain audio by relevant group and device. | Specialists set acceptance thresholds in advance. Until that decision, the app provides no automated pronunciation or reading-error score. |
| Does the child learn something that transfers? | Assess new words and texts outside the app at the beginning and end, with professional guidance. | A usability pilot does not support an effectiveness claim. Specify a comparison group, measurement tool and meaningful change for a follow-up study. |

Also measure practice frequency, parental help required and cost per completed session. Use these findings to determine the scope and affordability of the next version; downloads and time spent playing are not learning outcomes.

## Technical foundation and affordability

The existing React web app is the starting point. Any AI requests go through a server; API keys stay there. Manage costs with reviewed audio and stories prepared in advance, simple selection rules and speech processing only while a child is responding. Calculate audio minutes, generation, storage, hosting, content review, support and maintenance before proposing a price. Access through families or schools are models to investigate.

## Data and parental control

Request only data needed for the activity. A nickname and selected learning components may be sufficient; a full date of birth is not automatically necessary. Make clear to parents whether audio leaves the device, who processes it, why and for how long. Do not retain raw voice recordings by default; any research recordings require a separate, clear choice and a retention period.

For online services offered directly to children in Belgium, where data processing relies on consent, children under 13 cannot independently give that consent; consent from a parent or legal representative is required. The specific legal basis and any processing of health data must be assessed for the chosen implementation. See the [Belgian Data Protection Authority](https://www.gegevensbeschermingsautoriteit.be/professioneel/avg/rechtsgronden/toestemming).

## Current project status

At the initial review, the local version contained three activities: rhyming, blending sounds and reading words, with adult guidance and optional Dutch device speech. A pitch website and a fixed clickable reading experience are now being developed from this foundation. The delivery notes must separately identify which features actually work. This document describes the product plan and is not evidence that AI assessment, recording or storage have already been implemented.

The English presentation uses the name Reading Adventure, the mascot Fox and adapted examples. This presentation adaptation does not change the initial Dutch-speaking target audience or establish educational validation.

## Short pitch copy for the website

**Eyebrow:** A reading companion in development · For children aged 6–9

**Headline:** A little reading adventure. A step forward every day.

**Introduction:** Learning to read takes practice. Reading Adventure aims to give children short, playful reading moments, with clear support and room to try again. At home, alongside what your child learns at school.

**Primary button:** Try a reading adventure

**Secondary button:** Explore our plan

**Beside the demo:** Explore a sample session with words, a sentence and a question about the story. This clickable concept does not yet listen to or assess your child’s reading. The English demo presents an idea initially intended for Dutch-speaking children in Flanders.

**Three key points:**

- **Start small.** One activity at a time, with time to try.
- **Help within reach.** Listen, read together or take a short break.
- **See what was practised.** A simple recap for child and parent.

**AI section:** We are investigating how AI can choose suitable activities and support reading aloud. We begin with reviewed learning content and test speech feedback with specialists and families before children rely on it.

**Pronunciation section:** Pronunciation has its own development track, with exercises and assessment that we intend to develop with speech and language therapy expertise.

**Closing:** Our next goal: test whether children can practise comfortably, understand the support and move through a session more independently.

Do not use fictional testimonials, user counts, partners, proven learning gains or subscription prices. A contact or registration form belongs on the website only when a real recipient, processing and delivery have been configured.

## Evidence and limitations

- [Flemish guide to effective reading instruction](https://data-onderwijs.vlaanderen.be/documenten/bestand.ashx?nr=21349): supports explicit instruction and combining components of reading. The guide does not validate this app concept.
- [ASHA — Speech Sound Disorders: Articulation and Phonology](https://www.asha.org/practice-portal/clinical-topics/articulation-and-phonology/): background on different pronunciation difficulties and the importance of language and dialect context.
- [Carl et al., 2024 — Accuracy of Speech Sound Analysis](https://pubs.asha.org/doi/10.1044/2024_JSLHR-24-00009): research on one speech analysis system with children and adolescents speaking American English. Agreement was lower for incorrectly pronounced sounds. The results do not establish reliability for Dutch-speaking children.
- [Gao et al., 2024 — Speech recognition with children in the Netherlands](https://www.isca-archive.org/interspeech_2024/gao24d_interspeech.html): separate evaluation of transcription and reading-error detection; not validation for Flemish children aged 6–9.
- [Belgian Data Protection Authority — Consent](https://www.gegevensbeschermingsautoriteit.be/professioneel/avg/rechtsgronden/toestemming): the Belgian context for consent in online services for children.

The features, development sequence, session duration and pilot size above are design proposals. Reliability, learning gains, costs and willingness to pay still need to be investigated.