Description
Speech-to-text outputs can be evaluated with word error rate and related similarity measures. Researchers and accessibility developers use jiwer to compare transcripts against references. Reference text and transcripts may contain private speech, so evaluation data needs care.