JONAHKYAGABA/-Whisper-Amharic-ASR-Model-Evaluation

# 📢 Whisper Amharic ASR Model Evaluation # 📢 Whisper Amharic ASR Model Evaluation This project provides a complete pipeline for evaluating Whisper-based Automatic Speech Recognition (ASR) models on Amharic language datasets. It computes **Word Error Rate (WER)** and **Character Error Rate (CER)** across multiple model checkpoints and saves inference results to CSV for further analysis. --- ## 🎯 Objectives - Load pre-trained Whisper ASR models from Hugging Face - Run speech-to-text inference on Amharic audio datasets - Compute sentence-level and overall WER/CER scores - Export prediction results and logs in structured format --- ## 🧰 Dependencies Install required packages: ```bash pip install transformers datasets torchaudio jiwer pandas tqdm