A voice-AI system that listens to a child read aloud, encourages them like a patient saathi,
and generates observation-based fluency reports (WCPM, accuracy, error patterns) mapped to NIPUN Bharat लक्ष्य.
Live agent connectedDEMO — synthetic cohort dataGrade 1–3 · Hindi · आकलन "as, of & for" learning (NCF-FS 2022)
Children assessed128this demo cohort (synthetic)
Mean WCPM · Grade 352.4लक्ष्य: ≥ 60 शब्द प्रति मिनट
At grade लक्ष्य38%reading at/above NIPUN target
Mean accuracy91.2%words read correctly
Needs support21flagged for teacher attention
Live demo — बच्चा पढ़ता है, पठन साथी सुनती है
The child reads the passage below aloud. The AI saathi listens without interrupting, then gives warm,
specific encouragement — never marks, never scores aloud. Numeric scoring happens in the analysis pipeline, not in the conversation.
Reading passage · कक्षा 2–3 स्तर
55 शब्द · NIPUN Bharat स्तर के अनुरूप सरल कहानी
चिड़िया और दाना
एक छोटी चिड़िया थी। उसका नाम चीनू था। चीनू रोज़ सुबह दाना खोजने जाती थी।
एक दिन उसे आँगन में बहुत सारे दाने मिले। चीनू ने अपने दोस्तों को बुलाया।
सब चिड़ियों ने मिलकर दाना खाया। फिर सब ने मीठा गाना गाया।
शाम को चीनू अपने घोंसले में लौट आई। उसे बहुत अच्छी नींद आई।
पठन साथी से बात करें
Click the widget, allow the microphone, tell her your name and class, then read the story aloud.
Say “हो गया” when you finish.
Click the orb to start the call
पठन साथी identifies itself as an automated AI helper — never as a
person, teacher, or examiner. Demo content; reading scores are generated by the
offline analysis pipeline, not spoken in the call.
यह एक AI आवाज़ सहायक है, व्यक्ति नहीं — बातचीत की शुरुआत में यह स्वयं यह बताती है।
Word-level assessment report — sample child
Recorded audio → speech recognition → word-by-word alignment against the passage.
The report below is computed live on this page by the actual alignment algorithm, from one sample (synthetic) reading.
हर शब्द का रंग बताता है कि बच्चे ने उसे कैसे पढ़ा।
सही पढ़ा (correct) बदल कर पढ़ा (substitution) छोड़ दिया (omission)
Cohort view — कक्षा-वार प्रवाह वितरण
Distribution of WCPM across the assessed cohort, against NIPUN Bharat grade-level लक्ष्य bands (Grade 2 ≥ 45, Grade 3 ≥ 60 WCPM).
WCPM distribution · Grade 3 cohort
Synthetic demo data · n = 64
Children flagged for support
Sorted by WCPM · teacher follow-up list
Child
Grade
WCPM
Accuracy
Level
How the assessment works
STEP 1
बच्चा पढ़ता है
Child reads a calibrated grade-level passage aloud. The AI saathi greets, instructs, and listens without interrupting — warm, never test-like.
STEP 2
Speech recognition
Recorded audio is transcribed with word-level timestamps by a Hindi ASR model (Whisper / Bhashini), tuned for read-aloud speech.
STEP 3
Alignment & scoring
The transcript is aligned word-by-word against the reference passage (edit-distance alignment) → WCPM, accuracy %, omissions, substitutions.
STEP 4
Teacher report
Per-child and cohort reports mapped to NIPUN Bharat लक्ष्य. Children below band are flagged — a screening aid for the teacher, not a certificate.
Demonstration disclosure. All cohort numbers on this page are synthetic. The live agent gives qualitative
encouragement only; numeric WCPM scoring runs in the offline analysis pipeline. Child-speech ASR accuracy in Hindi is the
key validation area before field use — this system is proposed as a screening and practice aid under teacher
supervision, aligned with NCF-FS 2022 guidance on observation-based, stress-free assessment. AI disclosure is spoken in
every session. No real child data was used.