amkraneYoussef/TamazightTTS

FINE-TUNING SPEECHT5 FOR TAMAZIGHT TEXT-TO-SPEECH # Fine-Tuning SpeechT5 for Tamazight Text-to-Speech Companion Repository for: “Fine-Tuning SpeechT5 for Tamazight Text-to-Speech: A Foundation for Educational Technology in Low-Resource Languages” (Submitted to International Journal of Speech Technology) ## Overview This repository contains the resources, scripts, and configuration files used to fine-tune Microsoft’s SpeechT5 model for text-to-speech (TTS) synthesis in Tamazight, a low-resource language. The training pipeline uses Hugging Face Transformers and Datasets libraries, with detailed preprocessing, model training, and evaluation steps made publicly available. ## Quick Links - Preprocessed Training Set (HF-compatible): drive.google.com - Preprocessed Validation Set (HF-compatible): drive.google.com - Raw Audio Archive (with CSV files for train, val, and test): drive.google.com - Kaggle Preprocessing Notebook: kaggle.com - Training Pipeline (Colab): colab.research.google.com - Checkpoints + Logs + Best models: drive.google.com ## Dataset Sources This project was built using the following open-source speech datasets: | Dataset | Link | |-----------------------|---------------------------------------------------------------------| | Common Voice v13 | datacollective.mozillafound… | | tamazight_asr | TutlaytAI/tamazight_asr | | tamawalt-n-imZZyann | huggingface.co | …