Blog

Recent reviews of text-to-voice (AI) software options have discovered that when it comes to simply transcribing the spoken word into typed words, the software can do this with 98% accuracy.

But there is a lot more to an accurate transcript than getting the words right.

For example, voice-to-text software cannot always know which words to capitalise. Which words to italicse. Which words to place brackets around. Voice-to-text software cannot differentiate sounds (such as laugh, cry, sigh) from the spoken text. How to insert paragraphs in the best place. How to transcribe half-spoken words as well as fully spoken words. How to punctuation in such a way that ensures best readability while reflecting the speaker’s style of speaking.

That is, AI or voice-to-text software is not as accurate as a professional transcriptionist when it comes to producing accurate, readable transcripts.

The AI software you use to convert your voice to text may not have a  guarantee of confidentiality. In fact some software requires you to sign various rights away before using their software and does use the content provided by you to inform their software or future AI models.

Human transcriptionists do not do this. In fact, most human transcriptionists are happy to sign a guarantee of confidentiality. We at On Time Typing, Editing and Proofreading offer a signed guarantee of confidentiality to all our clients.

What a human’s guarantee of accuracy may include

We offer a signed guarantee of at least 98% accurate transcripts. We set our auto-backspace on longer than usual (for example, 2 seconds) so we are more or less proofreading our transcript as we go. Or, if the audio is not clear, we can firrst transcribe then proofread our transcripts against the audio to produce a transcript that is as accurate as possible, even when the audibility is inferior.

Our accurate transcripts include the following:

• A transcriber’s notes at the top of the transcript detailing the level of audibility, and accuracy, and showing a key with the identifier for each speaker
• correct capitalisation (including the client’s own preferences)
• person names spelled correctly (which may mean asking the client for the correct spellings – e.g. is it Smythe or Smith? Jane or Jayne?)
• place names spelled correctly (including a Google or glossary check)
• correct punctuation according to the spoken or oral style of the speaker – noting that if it is going to impact the style of speaking we do not attempt to correct it so it is grammatically correct. We ensure we keep the voice intact.
• paragraph breaks where the typist thinks it appropriate
• highlighting any words that may not be 100% correct due to inaudibility
• highlighting or italicising items that are “transcriber’s notes” such as a dictated ‘Author’s note to self for later on’; sounds such as Laughing, Crying, Sighing, Clapping, Papers rustling, etc.; or pauses (Pause.)

What voice-to-text or AI software can be used for

AI software can be used to produce transcripts of spoken content that you have recorded while  driving, or walking, or doing whatever, and you can then use those voice-to-text notes as reminders or as the basis for a report you are going to write, or whatever else your spoken content was about. It doesn’t  matter if the AI-produced punctuation is all over the place or if AI got a word wrong or edited what you said into something you didn’t mean. Like handwritten notes to self, a voice-to-text transcript of your spoken notes is just a rough copy of what you verbalised, and can be helpful.

The most accurate versions of voice-to-text software can be used to produce transcripts that suit your purpose as long as you do not need them to be perfectly verbatim – because AI cannot do that. And as long as you do not need a guarantee of confidentiality or accuracy – because AI cannot do that.

For example, a student doing a PhD may use Cockatoo or Rev to produce their transripts, as long as their institution does not require them to adhere to confidentiality or privacy rules, as long as absolute accuracy is not required.

For people who are not fast typists, it can be useful to get AI or voice-to-text software to produce a first draft transcript of a recording, which you can then use as your base transcript, and correct it against the audio yourself.

But if you are a fast typists, note that it takes the same time to correct (or proofread) an AI-produced transcript against the audio to produce an accurate transcript, as it does to produce that transcript from scratch.

That is, if you are a fast accurate typist and you need the end result to be an accurate transcript, then AI is not much use to you.

And what about AI and editing – including indexing?

I have yet to write an article comparing humans and AI when it comes to editing, but thankfully someone has recently written an excellent article about how useless AI is at indexing (which some editors specialise in). It is well-researched and as unbiased as its authors, who love indexes, could possibly be. Check it out: AI in indexing.

Image: thank you to Women in Tech Chat for permission to use this photograph.