Jolly Phonics is taught one child at a time. Early Day Academy asked whether it could be assessed that way too - for a whole class. Two multimodal modules: audio evaluation of sound production, image evaluation of letter formation.
evaluation parameters
Jolly Phonics teaches reading and writing through 42 letter sounds, each with a sound, an action and a letter formation. Assessment is one-to-one and oral - a teacher sits with a child, listens to them produce a sound, and watches them form a letter.
That is the right way to teach it, and an impossible way to measure it across a class. A teacher with thirty Early Years children cannot listen to each of them produce 42 sounds with enough attention to catch which phonemes are consistently wrong. By the time a pattern is noticed, months of practice have set it.
Handwriting carries the same problem in visual form. A teacher can see that letters are untidy without being able to say whether the cause is baseline drift, inconsistent stroke, or a letter formation that will later block cursive entirely.
Eight evaluation parameters: Phoneme Accuracy, Stress, Intonation, Rhythm, Clarity, Fluency, Target Words, Volume.
Audio is transcribed, analysed at phoneme level, then evaluated against the preset the teacher configured - carrying the target words and reference text for that particular session, rather than running a generic assessment.
Clickable timestamps. The module does not just return a score - it returns the moments. The teacher clicks a flag and the player jumps to the point where the child mispronounced. No teacher has to take the system's word for anything; the evidence is one click away.
Speech engines were selected against the constraint that actually matters here: this is children's speech, in Indian classrooms, with background noise.
| Engine | Role | Why |
|---|---|---|
| Speechace | Phonics assessment | Purpose-built for early literacy, scoring at syllable and phoneme level, with colour-coded letter feedback that never exposes phonetic notation to a five-year-old. |
| Microsoft Azure | Teacher-side reporting | Enterprise-grade scoring across accuracy, fluency, completeness and prosody, with batch processing for whole-class review. |
| SpeechSuper | Roadmap | Indian English dialect model for future phases. |
An engine trained only on native British or American English will mark a child down for an accent that is entirely correct where they live. Dialect is a design decision, not a detail.
Why we chose on more than accuracy
Eight evaluation parameters: Letter Formation, Size Consistency, Spacing, Baseline Alignment, Stroke Consistency, Slant, Legibility, Neatness.
The teacher uploads a reference image defining the target letter form, then batch-uploads student work. Every image returns a result per parameter rather than an overall neatness grade.
That breakdown is the entire value. "Untidy" is not something a teacher can act on. "Baseline alignment weak, letter formation sound" says the child knows the shape and needs ruled-line practice - a completely different intervention from a child whose formation is wrong.
Both modules follow one pattern - configure a preset, upload the work, receive evaluation and a report. A teacher who learns one can operate the other.
The sixteen parameters ship as sensible defaults and can be added to, removed, reordered and saved. The academy sets the standard; the system measures against it. Nothing is imposed by a rubric written elsewhere.
Both modules are fully specified - parameters, evaluation pipeline, reporting and speech engine selection complete - and developed in partnership with Early Day Academy under an active MOU.
The next phase is a classroom pilot with Early Years children at the academy.
Start with one department and one term, on your own rubric.
