Evaluate Speech Transcription Accuracy

Scenario

You are reviewing a speech-to-text model and need to explain how you would judge whether its transcripts are good enough for production use. The team wants a clear evaluation approach that covers both raw transcription quality and common failure cases in spoken audio.

Question

How do you evaluate the performance of a speech-to-text model, and what metrics would you use to measure transcription accuracy?