ahmednsalehm/quran-muaalem-tajweed

Quran Muaalem — Tajweed Grading Space (for Siraj)

Free ZeroGPU endpoint wrapping obadx/quran-muaalem (MIT). Grades a recitation against a specified āyah and returns per-letter tajweed/sifāt.

Deploy (needs a HuggingFace account — free)

1. 1.Create a new Space → SDK Gradio.
2. 2.Settings → Hardware → ZeroGPU (free; ~5 GPU-min/day on the free tier — plenty for solo testing).
3. 3.Upload app.py, requirements.txt, this README.md.
4. 4.Wait for build; first request downloads the ~660M model.

Call it from the Siraj Mushaf

The prototype's web/app/api/recite/verify/route.ts should POST the recorded audio + {sura, aya} to this Space (via @gradio/client or the Space /run/predict endpoint) instead of Gemini, then map the returned sifat[] diffs to inline corrections. web/components/MushafRecite.tsx already captures the audio.

Validation (2026-07-06)

Model proven on free Colab T4: 0.99s/clip · 70/70 sifāt match on Al-Husary (no false-positives) · wrong-āyah control correctly flagged (discriminates). See memory project_siraj_tajweed_from_mushaf.

⚠️ The ZeroGPU @spaces.GPU lazy-load in app.py hasn't been run on a live Space yet — test once on deploy.