🎙 LEHJA — build status
Updated 2026-06-13 12:54:01 UTC · auto-refresh 60s · cron every 5 min
OVERALL (Arabic track + MT): 28%
H100 NVL: 100% GPU · active jobs 9 · VRAM 33.7 GB used / 59.4 GB free of 93.6 GB
⛏ MINING LIVE — harvesting clean recitation clips with perfect tashkeel labels.
📋 Tasks — main & sub
✅ ✦. Urdu voice model (previous milestone) 100% ✅ CosyVoice2 Urdu LoRA trained + delivered · validation PASS · median CER 0.065 · model on app-server
✅ 1. Build Quranic-Arabic pipeline (canonical alignment, full tashkeel) 100% ✅ Diacritic-preserving aligner + mining/prep/train/validate scripts
✅ 2. Canonical Uthmani Quran text + alignment index 100% ✅ 6236 ayahs, full Uthmani Hafs (zabar/zer/pesh/shadda/tanwin/ghunna)
✅ Skeleton search index — rough transcript → exact ayah span
🔄 3. Mine in-domain Arabic recitation (your QTM corpus) 53% 🔄 whisper-ar VAD-segment + inline Quran align · 6,400/12,000 files · 0/8 shards done
🔄 clean recitation clips harvested · 6,409 clips · 695.1 min · 2,126 distinct ayahs
🔄 4. Canonical-align → diacritized labels + confidence gate 53% 🔄 perfect diacritized labels (alignment inline per clip)
⬜ high-confidence gate (match≥85) + dedup · 4,162 train-grade clips
⬜ 5. Train + validate + package Quranic Arabic LoRA 0% ⬜ CosyVoice2 fine-tune (learn diacritized recitation) · 25 epochs planned
⬜ validate: recites correct Arabic? (whisper-readback)
⬜ package CosyVoice2-0.5B-ar + download to app-server
⬜ MT. Flagship MT benchmark (Qwen3.5 vs qwen3:14b) 0% ⬜ latency + Urdu/Arabic quality on real tutor sentences
📊 Arabic data harvest
Recitation clips mined: 6,409 · 695.1 min clean audio
Train-grade (match≥85): 4,162 clips · 2,126 distinct ayahs covered
Source: your QTM Quran-recitation corpus (30,165 Arabic files available)
Approach: every clip's rough transcript is matched to the canonical Uthmani Quran text, so labels carry perfect diacritics (zabar/zer/pesh/tashdeed/tanwin/ghunna) — and non-recitation (Urdu explanation, repetition) is auto-filtered.