Datasetpublic

FaisaI/tadabur-align-references

tadabur-align-references Precomputed reference embeddings powering tadabur-align — word-level timestamp extraction for Quranic recitation via DTW alignment transfer (no ASR). What this is For 5,481 of the Quran's 6,236 ayahs, this dataset holds frame-level tadabur-embedding features for up to 8 reference reciters, plus each reference's word-level timestamps and internal-pause intervals. No audio is included — only model outputs and timing data. tadabur-align… See the full description on the dataset page: https://huggingface.co/datasets/FaisaI/tadabur-align-references.

sourceHugging Facecc-by-nc-4.0updated 2mo agoView on Hugging Face

0likes6.7kdownloads

Dataset Card

tadabur-align-references

Precomputed reference embeddings powering tadabur-align — word-level timestamp extraction for Quranic recitation via DTW alignment transfer (no ASR).

What this is

For 5,481 of the Quran's 6,236 ayahs, this dataset holds frame-level tadabur-embedding features for up to 8 reference reciters, plus each reference's word-level timestamps and internal-pause intervals. No audio is included — only model outputs and timing data. tadabur-align fetches these on demand instead of recomputing them locally.

Structure

S016A001/
  manifest.json              # word count, model hash, recipe version, reciter list
  Husary_64kbps.npz          # frame centers, 7 encoder layers (float16), word timestamps, pauses
  Alafasy_128kbps.npz
  ...                        # 4-8 files per ayah
quran-align/
  Husary_64kbps.json         # source word-timestamp data (see below)
  ...

Each .npz is versioned: it records the embedding-model hash and feature-recipe version it was built with, so a consumer can detect staleness after a model update rather than silently mixing incompatible embeddings.

Provenance

—Word timestamps: quran-align by cpfair (CC BY 4.0), bundled here under quran-align/.
—Reference audio: everyayah.com — used only to compute the embeddings in this dataset; not redistributed.
—Embeddings: FaisaI/tadabur-embedding.

License

CC BY-NC 4.0, matching the embedding model these features are derived from. Non-commercial use, with attribution to quran-align and everyayah for the underlying timing data and audio.