It’s Not FER

Generated by author using Midjourney

While Word Error Rate (WER) is the accepted standard for speech transcription, maybe Formatted Error Rate (FER) should be considered a better measure? FER takes into account all the additional parts of language that make for better readability and comprehension of a passage.

FER can also take into account paragraphs and other elements like capitalization. The good news is that with more language models becoming available, the ability to perform such formatting is being democratized.

Formatting accounts for a subtle and nuanced differences in the meaning of the text.

WORKS ON CONTINGENCY NO MONEY DOWN

Is different than…

Works on contingency? No, money down!

(thanks Lionel Hutz)

Maybe it’s time we give transcription the FER treatment?