Speech assessment for speech-therapy children in the Jordanian dialect

The problem

A speech therapist assesses a child's progress by ear and written notes, with no numeric measurement to compare sessions. Parents repeat the exercises at home without being able to check whether the child is pronouncing correctly, so mistakes can become ingrained. On top of that, all available tools are designed for English or Modern Standard Arabic while the child speaks the local dialect, so the tool rejects correct pronunciation and discourages the child.

How it works

1. DisplayThe device shows the word with a picture and a model pronunciation.
2. RegistrationA digital microphone captures the child's speech in high, noise-free quality.
3. AlignmentThe child's speech is aligned with the expected form, sound by sound.
4. EvaluationEach sound gets a score, with the type of error identified: substitution, omission, or distortion.
5. Showing the resultThe result is shown to the child in a simple, motivating way.
6. FollowingSession results build up into a progress curve the therapist can review.

Project features

• Trained on the Jordanian dialect — a gap the available tools haven't filled.
• It scores each sound within the word, instead of a single right-or-wrong verdict on the whole word.
• A progress curve across sessions lets the therapist follow improvement in numbers.
• It supports home practice: parents know immediately whether the pronunciation is right, so mistakes don't take root.
• It works offline, so children's voice data never leaves the device.

Expected challenges

These are the main obstacles to expect during the build, laid out clearly so you know them before you start.

• ⚠ The system is an aid for the therapist, not a diagnostic tool or a substitute for the therapy session.
• Children's voices differ from adults' — higher pitch and unstable pronunciation — so general models fail with them.
• Telling a pronunciation error from the natural features of the child's dialect is the hardest part of the project.
• Household noise hurts assessment accuracy, so the microphone's placement and isolation matter more than the model.
• Children's voice data is sensitive, so prior consent and local storage are basic requirements.

The consultation gives you a solution for each one.