Tajweed annotations for the Qur'an (riwayat hafs). The data is available as a JSON file with exact character indices for each rule, and as individual decision trees for each rule.
You can use this data to display the Qur'an with tajweed highlighting, refine models for Qur'anic speech recognition, or - if you enjoy decision trees - improve your own recitation.
The following tajweed rules are supported:
- Ghunnah (
ghunnah
) - Idghaam...
- With Ghunnah (
idghaam_ghunnah
) - Without Ghunnah (
idghaam_no_ghunnah
) - Mutajaanisain (
idghaam_mutajaanisain
) - Mutaqaaribain (
idghaam_mutaqaaribain
) - Shafawi (
idghaam_shafawi
)
- With Ghunnah (
- Ikhfa...
- Ikhfa (
ikhfa
) - Ikhfa Shafawi (
ikhfa_shafawi
)
- Ikhfa (
- Iqlab (
iqlab
) - Madd...
- Regular: 2 harakat (
madd_2
) - al-Aarid/al-Leen: 2, 4, 6 harakat (
madd_246
) - al-Muttasil: 4, 5 harakat (
madd_muttasil
) - al-Munfasil: 4, 5 harakat (
madd_munfasil
) - Laazim: 6 harakat (
madd_6
)
- Regular: 2 harakat (
- Qalqalah (
qalqalah
) - Hamzat al-Wasl (
hamzat_wasl
) - Lam al-Shamsiyyah (
lam_shamsiyyah
) - Silent (
silent
)
This project was built using information from ReciteQuran.com, the Dar al-Maarifah tajweed masaahif, and others.
All the data you probably need is in output/tajweed.hafs.uthmani-pause-sajdah.json
. It has the following schema:
[
{
"surah": 1,
"ayah": 1,
"annotations": [
{
"rule": "madd_6",
"start": 245,
"end": 247
},
...
]
},
...
]
The start
and end
indices of each annotation refer to the Unicode codepoint (not byte!) offset within the Tanzil.net Uthmani Qur'an text. Make sure to download the version with pause marks and sajdah signs, but without rub-el-hizb signs or me_quran tanween shapes. If you use different options or a different text entirely, you must rebuild the data file from scratch (at your own risk) - refer to the next section.
This data file is licensed under a Creative Commons Attribution 4.0 International License.
tajweed_classifier.py
is a script that takes Tanzil.net "Text (with aya numbers)"-style input via STDIN, and produces the tajweed JSON file (as described above) via STDOUT. It reads the decision trees from rule_trees/*.json
. Note that the trees have been built to function best with the Madani text; they rely on the prescence of pronunciation markers (e.g. maddah) that may not be present in other texts.
The following are renderings of the decision trees used to determine where each tajweed annotation starts and stops. Attributes are grouped by the letters they belong to, a letter being defined as a base character (e.g. ل) plus any diacritics that follow (codepoints in the Mn
category). Superscript/dagger alif is counted as a base character. The numbers prefixing each attribute indicate which letter the attribute belongs to: negative referring to previous letters, positive to future letters. Attributes starting with 0_...
refer to the exact character being considered. Annotations do not always start or stop on letter boundaries. Refer to tajweed_classifier.py
for the definition of each attribute.