JONAHKYAGABA/-Bemba-ASR-Evaluation-Pipeline

📢 Bemba ASR Evaluation Pipeline # 📢 Bemba ASR Evaluation Pipeline This project provides a complete pipeline for evaluating Automatic Speech Recognition (ASR) models on **Bemba-language** datasets. It focuses on calculating key transcription performance metrics — Word Error Rate (WER) and Character Error Rate (CER) — across multiple Hugging Face-hosted models. --- ## 🎯 Key Features - Loads and preprocesses Bemba test data (`KYAGABA/oromo_cleaned_dataset`) - Supports multiple models via Hugging Face `pipeline()` - Runs inference and logs predictions vs references - Computes: - Word Error Rate (WER) - Character Error Rate (CER) - Exports detailed CSV results for each model evaluated --- ## 🗂️ Dataset The test set is loaded from the Hugging Face Hub: