Datasets:

mrmuminov
/

quranic-corpus-morphology

dataset_info:
  features:
    - name: l_sura
      dtype: string
    - name: l_ayah
      dtype: string
    - name: l_word
      dtype: string
    - name: l_subword
      dtype: string
    - name: form
      dtype: string
    - name: tag
      dtype: string
    - name: features
      dtype: string
    - name: phoneme
      dtype: string
    - name: location
      list: string
  splits:
    - name: train
      num_bytes: 13752640
      num_examples: 128219
  download_size: 1763501
  dataset_size: 13752640
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
size_categories:
  - 100K<n<1M

Qur'an Phoneme + Harakat Dataset

This dataset contains phoneme-level and diacritic-level representations of the Qur'an text, based on the Quranic Arabic Corpus transliteration. It is intended for use in speech recognition and text-to-speech models, especially for phoneme-level fine-tuning of models like Whisper.

Dataset Structure

Column	Description
`LOCATION`	Quranic verse location in the format `(Sura:Ayah:Word:Subword)`
`FORM`	Original transliterated word form (Buckwalter-style)
`TAG`	Part-of-speech tag (e.g., N = Noun, P = Preposition, ADJ = Adjective)
`FEATURES`	Morphological features and lemma information
`PHONEME`	Phoneme-level representation with diacritics, separated by underscores and spaces

Example:

LOCATION	FORM	TAG	FEATURES	PHONEME
(1:1:1:1)	bi	P	PREFIX	bi+
(1:1:1:2)	somi	N	STEM	POS:N
(1:1:2:1)	{ll~ahi	PN	STEM	POS:PN

Usage

You can load the dataset directly from Hugging Face Hub:

from datasets import load_dataset

dataset = load_dataset("bahriddin/quran-phoneme-harakat", split="train")
print(dataset[0])