JONAHKYAGABA/oromo-ASR-evaluation-error-analysis-pipeline

Oromo ASR Evaluation & Error Analysis Pipeline This project provides a comprehensive framework for evaluating and analyzing the performance of automatic speech recognition (ASR) models on Oromo speech data. It includes WER/CER computation, error categorization, preprocessing enhancements, and vocabulary diagnostics. # oromo-ASR-evaluation-error-analysis-pipeline Oromo ASR Evaluation & Error Analysis Pipeline This project provides a comprehensive framework for evaluating and analyzing the performance of automatic speech recognition (ASR) models on Oromo speech data. It includes WER/CER computation, error categorization, preprocessing enhancements, and vocabulary diagnostics. Features 🧠 Model Evaluation using Hugging Face Transformers pipeline 📊 Metric Computation: Word Error Rate (WER), Character Error Rate (CER), substitutions, insertions, and deletions 🔁 Vocabulary Diagnostics: Preprocessing and remapping of problematic characters 🔍 Error Pattern Analysis: Frequent n-gram errors and contextual insights 📚 Training Data Comparison: Checks whether ASR errors appear in training vocabulary