dengjen-tashkeel
Arabic-text diacritic (tashkeel) restoration using an ONNX neural model, trained mainly on MSA data from Hareef.
Available as a Rust crate, a C ABI, a Python package, a Java library, and a standalone CLI.
Known limitation
Accuracy is bounded by the underlying Hareef model: it can mis-diacritize proper nouns and other
words whose correct case ending depends on context it wasn’t trained to resolve. For example,
given عن أمير المؤمنين أبي حفص عمر بن الخطاب, the name حفص comes out as حِفَصِ ("hifsi")
instead of the grammatically correct حَفْصٍ ("hafsin"). See
issue #28 for the full discrepancy. This
is a model-accuracy limitation, not a bug in this library’s code.
Credits
Created by mush42 (Musharraf Omer).
Want to help? Learn how to contribute to the ZirekHQ docs ›