Editing and proofing inaccurate auto speech-to-text transcripts is not recommended.
We regularly have requests to proof and edit transcripts that have been created from either speech-to-text programs or services such as Otter. There may possibly be an assumption that proofing and editing an inferior transcript is a relatively quick process but this is not the case, particularly for transcripts of interviews and focus groups.
The technology and accuracy of speech-to-text programs and services are improving all the time and it would be reasonable to expect that in not too many years’ time these may be the preferred option simply because of the cost effectiveness. In the case of a single speaker, or even two speakers, with exceptionally clear English, no accents and who are not prone to use idioms or slang, then the use of speech-to-text programs is possibly a reasonable choice. Of course that would eliminate most Australian speakers. We are a nation of many and varied accents but quite possibly one of the most difficult accents for auto speech-to-text to grapple with is the dinky-di Aussie.
Interviews and conversations present their own unique challenges. The tendency to over-speak and interrupt, the differences in speech patterns and accents all contribute to inaccuracies, and at times ludicrous guestimates, littered throughout the transcript making it unusable.
We have found that more often than not the editor has spent the same amount of time correcting an inaccurate transcript as it would have been to produce a fresh, accurate one. Editing auto speech-to-text transcripts is a tedious and lengthy process and for this reason we will no longer offer this service.
I cringe to think of clients editing the transcripts themselves because I would hope that your skills are better employed so if you have an inferior transcript that requires more than a cursory proof we will advise that you have us transcribe a fresh transcript.