Ayah Finder Lab — Quran WASM 0.4

Local Quran ASR • whisper.cpp • no ONNX

Recite what you remember.

The AI transcript is only a search hint. The Quran text shown as a result comes from the separate bundled corpus.

Diagnostic build. There is intentionally no caching service worker. Cloudflare Pages must honor the included _headers file so SharedArrayBuffer / WASM threads are available.

Phase A

1. Load whisper.cpp runtime

not loaded

This instantiates the high-memory WebAssembly engine by itself. It does not download the Quran model yet.

Ready to test runtime.

Phase B

2. Load Quran Whisper Base Q8

not loaded

Pinned Quran-tuned GGML Q8 model. Approx. 82 MB. Header is checked for the corrected 448-token decoder context before initialization.

Or load a GGML model file

Runtime required first.

Phase C

Recite / test audio

no audio

00:00.0

Model must initialize first.

Use audio file Diagnostic thread policy

Local matcher

Type, paste, dictate, or use AI transcript

—

Stress

Endurance

idle

No stress test run yet.

Device

Diagnostics

Raw engine output

Ayah Finder WASM 0.4