Datasets:

---

obadx

/

recitation-segmentation

like 1

This PR significantly enhances the dataset card for the recitation-segmentation dataset.

Key changes include:

• Updated Metadata:
  • Corrected size_categories to 100K<n<1M based on the "~300K annotated utterances" mentioned in the paper abstract.
  • Changed task_categories from voice-activity-detection to automatic-speech-recognition to better reflect the broader context of the paper (ASR-based pronunciation error detection and correction).
  • Added relevant tags (quran, arabic, tajweed, speech-segmentation) for improved discoverability.
• Enriched Content:
  • Added a detailed description of the dataset, highlighting its purpose (segmenting Quranic recitations using pause points for pronunciation error detection), construction methodology (98% automated pipeline), and scale (850+ hours of audio, ~300K utterances).
  • Included direct links to the associated research paper (https://huggingface.co/papers/2509.00094), the GitHub repository (https://github.com/obadx/recitations-segmenter), and the project page (https://obadx.github.io/prepare-quran-dataset/).
  • Added a "Sample Usage" section with Python code directly from the GitHub README, demonstrating how to load and process audio for segmentation using the recitations-segmenter library.

These improvements make the dataset more informative, discoverable, and easier to use for the Hugging Face community.