Build a caption QA harness in Python: WER, missed entities, timing and reading rate

TL;DR We're building a caption evaluation harness that scores a WebVTT file on four axes...