BounharAbdelaziz/Youtube-Commons-Moroccan-Darija

# YouTube Commons Project for Moroccan Darija This project automates the download and transcription of Creative Commons (CC-BY) licensed videos from selected Moroccan YouTube channels. The audio content is transcribed using Gemini with support for code-switching between Darija, French, and English. Proper credit is given to all content creators. Additionally, this repository includes code for training and evaluating speech-to-text (STT) models for Moroccan Darija, along with a Hugging Face Space for testing them interactively. ## Features - **Automated Download**: Fetches CC-BY licensed videos only. - **Accurate Transcription**: Uses Gemini for multi-language transcription. - **Content Crediting**: Ensures content creators are credited for their work. - **STT Model Training & Evaluation**: Fine-tune and test STT models on Moroccan Darija. - **Hugging Face Space**: Provides an interactive interface to test the STT models. ## Speech-to-Text (STT) Models This repository also includes code for fine-tuning and evaluating speech-to-text models on Moroccan Darija. The models are trained using Whisper and other transformer-based architectures. The goal is to improve transcription accuracy for Moroccan Darija, including code-switching scenarios. ## Test the STT Models A Hugging Face Space is available for testing the trained STT models interactively. You can upload your own audio clips and evaluate transcription quality in real-time! ## Requirements Install dependencies: ```bash pip install -r requirements.txt ``` ## Usage ### Data Collection Run the script to process all listed channels: ```bash python data_collection/youtube_cc_by_scraper.py ``` ### STT Model Training Collection Run the script to process all listed channels: ```bash python train_model/main.py ``` ## YouTube Channels List | Channel URL | Description | CC-BY | Scraped | |-------------------------------------------- …