Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Deep learning for AR text Vocalization - التشكيل الالي للنصوص العربية
| Date | Stars |
|---|---|
| 2026-07-31 | 355 |
| 2026-08-06 | 355 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Shakkala Project مشروع شكّالة
[](https://badge.fury.io/py/shakkala)
<img src="images/shakkala.png" alt="Model" height="140" width="140"/>
## Introduction
The Shakkala project presents a recurrent neural network for Arabic text vocalization that automatically forms Arabic characters (تشكيل الحروف) to enhance text-to-speech systems. The model can also be used in other applications such as improving search results. In the beta version, the model was trained on over a million sentences, including a majority of historical Arabic data from books and some modern data from the internet. The accuracy of the model reached up to 95%, and in some data sets it achieved even higher levels of accuracy depending on complexity and data distribution. This innovative approach has the potential to significantly improve the quality of writing and text-to-speech systems for the Arabic language.
## Requirements
### Easy setup
No GitHub repository installation is needed for [pip](https://pypi.org/project/shakkala/) case:
```
pip install shakkala
```
### Project setup
Execute the source code from Github:<br/>
```
cd requirements
pip install -r requirements.txt
cd ..
```
Note: Shakkala has been tested with Tensorflow 2.9.3.<br>
## Code Examples (How to)
Check full example in (demo.py) file.<br/>
0. Import
```
from shakkala import Shakkala
```
1. Create Shakkala object
```
sh = Shakkala()
```
OR for advanced usage:
```
sh = Shakkala(version={version_num})
```
2. Prepare input
```
input_text = "فإن لم يكونا كذلك أتى بما يقتضيه الحال وهذا أولى"
input_int = sh.prepare_input(input_text)
```
3. Call the neural network
```
model, graph = sh.get_model()
logits = model.predict(input_int)[0]
```
4. Predict output
```
predicted_harakat = sh.logits_to_text(logits)
final_output = sh.get_final_text(input_text, predicted_harakat)
```
Available models: <br>
- version_num=1: First test of the solution.
- version_num=2: Main release version.
- version_num=3: Some enhancements from version number 2.
It worth to try both version_num=2 and version_num=3.
### Demo run
The fastest way to start with Shakkala by running the demo from Github:
```
python demo.py
```
## Perfomance Tips
Shakkala built in object oriented way to load the model once into memory for faster prediction, to make sure you dont load it multiple times in your service or application follow the steps:
- Load the model in global variable:
```
sh = Shakkala(folder_location, version={version_num})
model, graph = sh.get_model()
```
- Then inside your request function or loop add:
```
input_int = sh.prepare_input(input_text)
logits = model.predict(input_int)[0]
predicted_harakat = sh.logits_to_text(logits)
final_output = sh.get_final_text(input_text, predicted_harakat)
```
## Accuracy
In this beta version 2 accuracy reached up to 95% and in some data it reach more based on complexity and data disribution.
This beta version trained on more than million sentences with majority of historical Arabic data from books and **some of** available formed modern data in the internet.<br/>
<img src="https://github.com/Barqawiz/Shakkala/blob/master/images/loss_history_v2.png" alt="history" style="height: 350px;"/>
### Prediction Example
For live demo based on Shakkala library click the [link](http://ahmadai.com/shakkala/) <br/>
| Real output | Predicted output |
| ------------- | ---------------- |
| فَإِنْ لَمْ يَكُونَا كَذَلِكَ أَتَى بِمَا يَقْتَضِيهِ الْحَالُ وَهَذَا أَوْلَى | فَإِنْ لَمْ يَكُونَا كَذَلِكَ أَتَى بِمَا يَقْتَضِيهِ الْحَالُ وَهَذَا أَوْلَى |
| قَالَ الْإِسْنَوِيُّ وَسَوَاءٌ فِيمَا قَالُوهُ مَاتَ فِي حَيَاةِ أَبَوَيْهِ أَمْ لَا | قَالَ الْإِسْنَوِيُّ وَسَوَاءٌ فِيمَا قَالُوهُ مَاتَ فِي حَيَاةِ أَبَوَيْهِ أَمْ لَا |
| طَابِعَةٌ ثُلَاثِيَّةُ الْأَبْعَاد | طَابِعَةٌ ثَلَاثِيَّةُ الْأَبْعَادِ |
### Accuracy Enhancements
The model can be enhanced to reach more than 95% accuracy with following:<br/>
- Availability of more formed **moderExcerpt of 5,987 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:51ac79988daddff6, topic:deep-learning