JONAHKYAGABA/Kinyarwanda-Speech-Translation-Evaluation-Framework

Kinyarwanda Speech Translation Evaluation Framework # 🇷🇼 Kinyarwanda Speech Translation Evaluation Framework This project implements a **qualitative evaluation pipeline** for Kinyarwanda speech translation outputs. It uses **textual references and hypothesis files** to compute a range of NLP evaluation metrics including BLEU, CHRF, WER, CER, and ROUGE. --- ## 🎯 Objectives - Evaluate **machine-generated Kinyarwanda translations** - Compare system predictions with gold references - Compute standardized NLP metrics for quality assessment - Support batch evaluation from CSV datasets --- ## 🔍 Key Features - ✅ BLEU and CHRF (n-gram overlap) - ✅ WER and CER (edit distance-based metrics) - ✅ ROUGE-1, ROUGE-2, ROUGE-L (recall-oriented summarization metrics) - ✅ Command-line driven evaluation - ✅ CSV output of all metric scores for easy analysis --- ## 🧰 Requirements Install the required packages: ```bash pip install evaluate pandas numpy sacrebleu rouge_score jiwer