Team Ai
Datasetpublic

Mathani-Ayat/qaloon-reciter-experiments

Experimental only — Waleed and Trabulsi Private research inventory uploaded at the project owner's request. Not part of the approved training dataset. Both sources retain authorization_status: unauthorized; an experimental upload is not source redistribution permission. Waleed: 569 individually labelled WAVs, durations read from WAV headers; Fātiḥah uses a separate basmalah track and fused ending. Source labels have not been independently checked. Some source identity entries… See the full description on the dataset page: https://huggingface.co/datasets/Mathani-Ayat/qaloon-reciter-experiments.

sourceHugging Faceotherupdated 8d agoView on Hugging Face
0likes17downloads
Dataset Card

Experimental only — Waleed and Trabulsi

Private research inventory uploaded at the project owner's request. Not part of the approved training dataset. Both sources retain authorization_status: unauthorized; an experimental upload is not source redistribution permission.

  • —Waleed: 569 individually labelled WAVs, durations read from WAV headers; Fātiḥah uses a separate basmalah track and fused ending. Source labels have not been independently checked. Some source identity entries say cached; use the reciter_key instead of inferring identity from that string.
  • —Trabulsi: 603 source WAVs, converted individually to 16 kHz mono PCM16. Original source-code identity is retained. 35 clips belong to a partial al-Baqarah capture; this inventory extends beyond the current MVP references. metadata.json and ANALYSIS.md contain coverage/gap/fusion candidates. Nine expected-code gaps and 43 review candidates are not confirmed acoustic omissions/fusions. Machine transcription hints require listening review. Pending Trabulsi clips have no trusted training label.
  • —No raw MP3s, signed captured URLs, local conversion sidecars or credentials are included. Saltany remains outside the release.

All rows carry release_type: experimental_only and usable_for_training: false. AudioFolder's train config split is an inventory view, not a sanctioned training partition. Keep these rows out of automatic production/baseline dataset loaders. Do not merge metadata or assume source file IDs are corrected Qālūn ayah IDs.

Use the reviewer workflow from the project src/hub_release/README.md to confirm single/fused/basmalah-only clips. Record complete combined text for fusions; do not trim fixed basmalah durations or claim a filename gap is a missing spoken ayah. Authorization and alignment review are separate fields.

Release manifest stores content SHA-256 hashes and version. Pin a specific Hub commit for experiments and preserve experimental provenance in any run. See LICENSE: this is not an open audio license.