WER
Word Error Rate. Percentage of words that are wrong compared to edited ground truth. Lower is better.
Model Evaluation
Training Info
This page is the hub for model-vs-model comparison reports. Use it to evaluate whether a new training run actually improves scanner transcription quality before promoting it.
CER
Character Error Rate. Useful when numbers, unit IDs, and short tokens matter. Lower is better.
Delta
Difference between model metrics. Negative delta for Model B means B improved over A.
Per-Call Wins
Shows where a model performs better call-by-call, not just in aggregate averages.
Training Output Reports
Select a report to view its summary and call grid below.
Report: whisper_medium_v100_scan_vs_whisper_medium_v101_scan_20260419_111337_training_result_set.html
← Scroll horizontally inside the report to see all model columns →