quranic-phonemizer 2.0

A Grapheme-to-Phoneme converter (G2P) for the Qurʾan (Hafs riwaya), converting text to phoneme sequences with comprehensive support for all tajweed rules and waqf phonetic effects.

Qurʾanic Phonemizer

A Grapheme-to-Phoneme converter (G2P) for the Qurʾan (Hafs riwaya), converting text to phoneme sequences with comprehensive support for all tajweed rules and waqf phonetic effects.

Potential use cases:

• Speech Recognition: Phonetically transcribe recitations, create training data for machine learning systems
• Text-to-Speech: Develop accurate TTS systems for Qurʾanic Arabic
• Linguistic & Tajweed Analysis: Study phonological patterns and tajweed rule distributions across the Qurʾan, apply tajweed rule labels and coloring
• Educational Tools: Build interactive applications for assessing Qur'an and tajweed pronunciation
• Timing Analysis: Generate word-by-word timestamps for recitations, analyse madd/ghunnah durations

In addition to the Python API, the phonemizer can be used interactively: quranicphonemizer.com.

Table of Contents

• Phoneme Inventory
• Usage
• Input References
• Text Search
• Outputs
• Stops (Waqf)
• Phonetic Text
• Tajweed Mappings
• Letter-Phoneme Mappings
• Contributing
• Credits
• Citing

Phoneme Inventory

The phoneme inventory uses the standard International Phonetic Alphabet (IPA) Arabic phonemes alongside custom phonemes for Tajweed rules, totalling 69-71 phonemes (depending on Tajweed configuration).

All phonemes are configurable in resources/base_phonemes.yaml and resources/rule_phonemes.yaml.

Consonants

Letter	Phoneme	Letter	Phoneme	Letter	Phoneme	Letter	Phoneme
أ , إ , ء , ؤ , ئ	ʔ	د	d / dd	ض	dˤ / dˤdˤ	ك	k / kk
ب	b / bb	ذ	ð / ðð	ط	tˤ / tˤtˤ	ل	l / ll / lˤlˤ
ت	t / tt	ر	r / rˤ / rr / rˤrˤ	ظ	ðˤ / ðˤðˤ	م	m
ث	θ / θθ	ز	z / zz	ع	ʕ / ʕʕ	ن	n
ج	ʒ / ʒʒ	س	s / ss	غ	ɣ	هـ	h / hh
ح	ħ / ħħ	ش	ʃ / ʃʃ	ف	f / ff	و	w / ww
خ	x / xx	ص	sˤ / sˤsˤ	ق	q / qq	ي , ى	j / jj

Gemination (shaddah) is represented by repeating the phoneme to create new distinct phonemes. Note that there is no gemination for m / n (modelled as tajweed instead), and for ʔ / ɣ (do not exist in the Qurʾān).

Vowels

Vowel	Phoneme
َ	a / aˤ
ُ	u
ِ	i
ا , ى	a: / aˤ:
و	u:
ي , ى	i:

Tajweed Rules

Rule	Phoneme
Iqlab	ŋ
Idgham	ñ / m̃ / j̃ / w̃
Ikhfaa	ŋ (Light) ŋˤ (Heavy) ŋ (Shafawi)
Qalqala	Q (Sughra) QQ (Kubra)
Tafkheem	lˤlˤ (Lam in "Allah") rˤ / rˤrˤ (Raa)

Usage

Installation

pip install quranic-phonemizer

Quick Start

from quranic_phonemizer import Phonemizer

pm = Phonemizer()
res = pm.phonemize("1:1")
print(res.text())
print(res.phonemes_str())

بِسْمِ ٱللَّهِ ٱلرَّحْمَـٰنِ ٱلرَّحِيمِ ١

bismi lla:hi rˤrˤaˤħma:ni rˤrˤaˤħi:m

Input References

phonemize() accepts a variety of flexible formats to specify which part of the Qurʾān to phonemize:

Format Example	Meaning
"1"	Entire chapter 1
"1:1"	Verse 1 of chapter 1
"1:1:1"	Word 1 of verse 1 of chapter 1
"1:1 - 1:4"	Verse range: 1:1 through 1:4
"1:1 - 1:2:2"	From 1:1 to word 2 of 1:2
"1 - 2:2"	From entire chapter 1 through verse 2 of chapter 2

Text Search

Instead of a reference, you can pass Arabic text directly using ref_text to fuzzy-match against the Uthmanic Hafs text of the Qur'an:

res = pm.phonemize(ref_text="بسم الله الرحمن الرحيم")
print(res.ref)
print(res.match_score)
print(res.phonemes_str())

1:1:1-1:1:4

0.903

bismi lla:hi rˤrˤaˤħma:ni rˤrˤaˤħi:m

The match_score attribute (0–1) indicates how closely the input text matched the Qurʾānic text. You can also scope the search to a specific surah or range by combining ref and ref_text:

res = pm.phonemize(ref="2", ref_text="الله لا إله إلا هو الحي القيوم")
print(res.ref)
print(res.match_score)
print(res.phonemes_str())

2:255:1-2:255:7

0.836

ʔalˤlˤaˤ:hu la: ʔila:ha ʔilla: huwa lħajju lqaˤjju:m

Outputs

phonemize() returns a PhonemizeResult object, containing:

Attribute	Description
ref	The resolved reference string
match_score	Fuzzy match confidence (0–1) when using ref_text; None otherwise
text()	The Qurʾānic text
phonemes_list(split)	Phoneme lists grouped by split: "word", "verse", or "both"
phonemes_str(phoneme_sep, word_sep, verse_sep)	Full phoneme string, configurable with separators
show_table(phoneme_sep, split)	Pandas DataFrame view, grouped by split (requires pandas)
save(path, *, fmt, split)	Save results to JSON, CSV, or mapping format
phonetic_text(word_sep, verse_sep)	Recitation-accurate display text with stopping/starting transforms

Output Example (Phonemes String)

res = pm.phonemize("112", stop_signs=["verse"])
print(res.text())
print(res.phonemes_str(phoneme_sep=" ", word_sep=" | ", verse_sep="\n"))

قُلْ هُوَ ٱللَّهُ أَحَدٌ ١ ٱللَّهُ ٱلصَّمَدُ ٢ لَمْ يَلِدْ وَلَمْ يُولَدْ ٣ وَلَمْ يَكُن لَّهُۥ كُفُوًا أَحَدٌ ٤

q u l | h u w a | lˤlˤ aˤ: h u | ʔ a ħ a d Q | ʔ a lˤlˤ aˤ: h u | sˤsˤ aˤ m a d Q | l a m | j a l i d Q | w a l a m | j u: l a d Q | w a l a m | j a k u | ll a h u: | k u f u w a n | ʔ a ħ a d Q

Stops (Waqf)

Optionally, pass stop_signs=[] to apply stops at Quranic stop signs, and/or stop_refs=[] to stop at specific word locations:

Stop key	Symbol
"verse"	۝
"preferred_continue"	ۖ
"preferred_stop"	ۗ
"optional_stop"	ۚ
"compulsory_stop"	ۘ
"prohibited_stop"	ۙ

ref = "68:33"
res = pm.phonemize(ref)
print(res.text())
print(res.phonemes_str())

res = pm.phonemize(ref, stop_signs=["preferred_continue"])
print(res.phonemes_str())

res = pm.phonemize(ref, stop_signs=["optional_stop"])
print(res.phonemes_str())

كَذَٰلِكَ ٱلْعَذَابُ ۖ وَلَعَذَابُ ٱلْـَٔاخِرَةِ أَكْبَرُ ۚ لَوْ كَانُوا۟ يَعْلَمُونَ ٣٣

kaða:lika lʕaða:bu walaʕaða:bu lʔa:xirˤaˤti ʔakbarˤu law ka:nu: jaʕlamu:n

kaða:lika lʕaða:bQ walaʕaða:bu lʔa:xirˤaˤti ʔakbarˤu law ka:nu: jaʕlamu:n

kaða:lika lʕaða:bu walaʕaða:bu lʔa:xirˤaˤti ʔakbarˤ law ka:nu: jaʕlamu:n

# Stop at a specific word location
res = pm.phonemize("1:1-1:3", stop_refs=["1:2:2"])

Phonetic Text

phonetic_text() returns a recitation-accurate rendering of the Arabic text, applying the phonetic transforms that occur when starting or stopping on a word. This is useful for displaying text as it would actually be recited.

Starting Transforms

Transform	Original	Phonetic Text
Hamza wasl → fatha	ٱلرَّحْمَٰنِ	أَلرَّحْمَٰنْ
Hamza wasl → damma	ٱدْعُ	أُدْعْ
Hamza wasl → kasra	ٱهْدِنَا	إِهْدِنَا
Remove first-letter shaddah	لِّلْمُتَّقِينَ	لِلْمُتَّقِينْ

Stopping Transforms

Transform	Original	Phonetic Text
Haraka → sukun	ٱلرَّحِيمِ	ٱلرَّحِيمْ
Taa marbuta → haa + sukun	رَحْمَةِ	رَحْمَهْ
Madd iwad (alef)	كِتَٰبًا	كِتَٰبَا
Madd iwad (hamza)	دُعَآءً	دُعَآءَا
Strip madd silah	حَوْلَهُۥ	حَوْلَهْ

Other Transforms

Transform	Original	Phonetic Text
Allah dagger alef	ٱللَّهِ	ٱللَّـٰهِ

Huroof Muqattaʿat

Opening letters are returned as their spelled-out recitation forms:

Text	Phonetic Text
الٓمٓ	أَلِفْ لَآم مِّيٓمْ
الٓر	أَلِفْ لَآمْ رَا
الٓمٓصٓ	أَلِفْ لَآم مِّيٓمْ صَآدْ
الٓمٓر	أَلِفْ لَآم مِّيٓمْ رَا
كٓهيعٓصٓ	كآفْ هَا يَا عَيْن صَآدْ
طه	طَا هَا
طسٓمٓ	طَا سِيٓن مِّيٓمْ
طسٓ	طَا سِيٓنْ
يسٓ	يَا سِيٓنْ
صٓ	صَآدْ
حمٓ	حَا مِيٓمْ
عٓسٓقٓ	عَيْن سِيٓن قَآفْ
قٓ	قَآفْ
نٓ	نُوٓنْ

Tajweed Mappings

tajweed_mappings() returns per-letter tajweed rule annotations for any phonemized passage. Each Arabic letter is annotated with the rules it participates in, distinguishing between source rules (rules the letter triggers) and target rules (rules affecting this letter from another letter). Annotations account for starting and stopping effects — cross-word rules disappear when stopping, while rules like qalqala_kubra and madd_arid_lissukun only appear at stops.

result = pm.phonemize("1:1", stop_signs=["verse"])
tajweed = result.tajweed_mappings()
print(tajweed.to_json(indent=2))

Example output for ٱلرَّحْمَـٰنِ (continuing):

{"location": "1:1:3", "entries": [
  {"char": "ٱ", "source_rules": ["hamza_wasl_silent"]},
  {"char": "ل", "source_rules": ["lam_shamsiyah"]},
  {"char": "ر", "source_rules": ["tafkheem"], "target_rules": ["lam_shamsiyah"]},
  {"char": "ح"},
  {"char": "م"},
  {"char": "ٰ", "source_rules": ["madd_tabii"]},
  {"char": "ن"}
]}

Extension characters (dagger alef ٰ, mini waw ۥ, mini yaa ۦ) are split into their own entries so their madd rules are kept separate from the base letter. Huroof muqattaat are returned in their spelled-out recitation form (e.g. الٓمٓ → أَلِفْ · لَآم · مِّيٓمْ).

Rules

Source-only — rules that annotate only the letter itself:

• tafkheem
• noon_ghunnah, meem_ghunnah
• qalqala_sughra, qalqala_kubra
• vowel_silent, silent_iltiqaa_sakinayn, iltiqaa_sakinayn_tanween
• hamza_wasl_silent, hamza_wasl_fatha, hamza_wasl_kasra, hamza_wasl_damma
• madd_tabii, madd_wajib_muttasil, madd_jaiz_munfasil, madd_lazim, madd_arid_lissukun, madd_leen

Source + target — the source letter triggers the rule and a second letter is annotated as the target:

• iqlab_noon, iqlab_tanween
• ikhfaa_noon, ikhfaa_tanween, ikhfaa_shafawi
• idgham_ghunnah_noon, idgham_ghunnah_tanween, idgham_shafawi
• idgham_bila_ghunnah_noon, idgham_bila_ghunnah_tanween
• idgham_mutamathilayn, idgham_mutaqaribayn, idgham_mutajanisayn_kamil, idgham_mutajanisayn_naqis, lam_shamsiyah

For full details, examples, and multi-rule overlap documentation, see docs/tajweed-mappings.md.

Letter-Phoneme Mappings

letter_phoneme_mappings() returns flat [chars, phonemes] pairs where every entry has at least one phoneme. Silent letters are merged into adjacent entries rather than appearing with empty phonemes, and word boundaries are encoded as spaces in the chars field.

result = pm.phonemize("1:1")
lpm = result.letter_phoneme_mappings()
for chars, phonemes in lpm.to_list():
    print(f"{chars!r} -> {phonemes}")

'ب' -> ['b', 'i']
'س' -> ['s']
'م ' -> ['m', 'i']
'ٱلل' -> ['ll', 'a:']
'ه ' -> ['h', 'i']
'ٱلر' -> ['rˤrˤ', 'aˤ']
'ح' -> ['ħ']
'م' -> ['m']
'ٰ' -> ['a:']
'ن ' -> ['n', 'i']
'ٱلر' -> ['rˤrˤ', 'aˤ']
'ح' -> ['ħ']
'ي' -> ['i:']
'م' -> ['m']

Merge Rules

Silent letters merge in one of three directions:

Direction	When	Example
PREV	Silent vowel letter merges into previous entry	"وا" -> ['w'] — silent alef appended to waw
NEXT	Silent letter at word start merges into next entry	"ٱلر" -> ['rˤrˤ', 'aˤ'] — hamza wasl + lam into raa
CROSS-WORD	Silent letter at word end merges with next word's first	"ن ر" -> ['rˤrˤ', 'aˤ'] — space inside chars

When both sides of a word boundary have phonemes, they stay separate with a space suffix on the last entry: "ن " -> ['ŋ'].

Extension characters (dagger alef, mini waw, mini yaa) are split into their own entries. Mappings reflect stopping/starting context — entry count and merge patterns change depending on waqf.

For full details, merge rule reference, and validation rules, see docs/letter-phoneme-mappings.md.

Contributing

If you find any issues or have feature suggestions, please feel free to open an issue or submit a pull request.

Future plans include detailed tajweed annotations and support for other turuq and riwayat.

Credits

The project makes use of the Quranic Universal Library's (QUL) Hafs script.

Citing

If you use this phonemizer in your work, please cite the paper as follows:

@inproceedings{
ibrahim2025quranic,
title={Qur{\textquoteright}anic Phonemizer: Bringing Tajweed-Aware Phonemes to Qur{\textquoteright}anic Machine Learning},
author={Ahmed Ibrahim},
booktitle={5th Muslims in ML Workshop co-located with NeurIPS 2025},
year={2025},
url={https://openreview.net/forum?id=hZt0JK28iV}
}