AnasGT/text-to-speech_Algerian_Dailect

Text-to-Speech (TTS) system for the Algerian dialect using AI and deep learning techniques. # Egyptian-Algerian Text-to-Speech (XTTS) Fine-Tuning Project ## 🎯 Project Overview This is a final-year project that implements a complete pipeline for building a specialized text-to-speech model for **Algerian Arabic dialect**. The project fine-tunes a pre-trained XTTS model on Algerian data through a **two-stage iterative approach**: first with a small YouTube dataset for initial adaptation, then with a larger dataset and improved preprocessing pipeline for production-quality results. ### 📌 Objective 1. Build a specialized TTS model for Algerian Arabic dialect 2. Collect and preprocess Algerian audio data from YouTube (two stages) 3. Fine-tune model iteratively: small data → large data with improved preprocessing 4. Achieve production-quality speech synthesis for Algerian speakers 5. Provide GUI application for speech synthesis and playback with subtitles --- ## 📁 Project Structure ### Root Level ``` L3_PFE/ ├── README.md # Project documentation ├── Alasmar_Anas_last_v.pdf # Project report/thesis ├── get_data/ # Data collection module ├── data_preprocessing/ # Data preprocessing and cleaning ├── finetuning/ # Model fine-tuning notebooks ├── inference/ # Inference and testing modules └── app/ # Desktop application (PySide GUI) ``` --- ## 📊 Data Flow Pipeline ``` XTTS Base Model ↓ YouTube (Algerian Content) ↓ [get_data/] → Download Algerian audio (Stage 1 & 2) ↓ Algerian Audio + Transcripts ↓ [data_preprocessing/] → Clean, validate, add emotions & phonetics → Stage 1: Basic preprocessing → Stage 2: Improved preprocessing ↓ Processed Algerian Dataset (Excel files) ↓ [finetuning/] → Fine-tune model (Stage 1 & 2) → Stage 1: Small data, basic preprocessing → Stage 2: Large data, improved prepr …