لهجتنا — Arabic Dialect Text-to-Speech Model
Model Summary
لهجتنا is an open Arabic Text-to-Speech (TTS) model designed to generate natural-sounding speech across Arabic dialects.
The goal of لهجتنا is to create a unified speech model capable of representing spoken Arabic dialects from across the Arab world, capturing their phonetic diversity, rhythm, and prosody.
Unlike many Arabic TTS systems that primarily focus on Modern Standard Arabic (MSA), this model is designed to synthesize real conversational dialect speech.
The model supports Arabic text with diacritics (تشكيل) to improve pronunciation accuracy and speech naturalness.
Model Details
Model Name: لهجتنا
Task: Text-to-Speech (TTS)
Language: Arabic Dialects
Architecture: Based on the Chatterbox Multilingual TTS architecture
Supported Dialects
The current version of the model includes support for several Arabic dialects, with additional dialects planned as the project evolves.
Dialect Coverage Roadmap
The long-term goal of لهجتنا is to support all Arabic dialects within a single unified model.
Progress will be tracked using the checklist below.
- Egypt
- Saudi Arabia
- Morocco
- Iraq
- Sudan
- Palestine
- Lebanon
- Syria
- Libya
- Tunisia
- United Arab Emirates
- Kuwait
- Qatar
- Bahrain
- Oman
- Yemen
- Jordan
- Algeria
- Mauritania
These checkboxes will be updated as dialect support improves and new datasets are incorporated.
Known Limitations
Repetition Issue
In some cases, the model may generate repeated words or phrases during speech generation.
This behavior can usually be controlled by adjusting the repetition_penalty parameter during inference.
Increasing the repetition penalty can help reduce repetitive outputs and produce more stable speech generation.