Accurate, or actually finished

Typing every word yourself produces the best result on any single clip and stops being possible somewhere around the fifth one.

01

What is the practical arrangement?

Automatic first, corrected second. Automatic handles the ordinary words, which is the overwhelming majority, and the corrections concentrate in a small set: names, technical terms, figures and anything said unclearly. Reading those before producing gets close to typed accuracy for a fraction of the effort.

02

Where corrections concentrate

Reading only the four rows below the first gets close to typed accuracy in a fraction of the time, which is what makes it a check you will actually do.

Word typeShare of a clipShare of the errors
Ordinary vocabularyMost of itVery few
Proper nounsA handfulA large share
Technical termsA handfulA large share
NumbersA fewSome, and the costliest
Words spoken over another voiceA few secondsA large share

03

Building a list of words you always correct

After a few clips you will notice the same corrections again and again: a guest's surname, your product name, a piece of industry jargon. Write them in a short list and keep it next to you while reviewing. Checking a known list is faster than reading every line with fresh eyes.

Share that list with anyone else who reviews your clips. It captures knowledge that otherwise lives in one person's head, and it makes the result consistent when several people handle the same show.

Review your list every few months. Some entries stop mattering when a guest leaves the show or a product is renamed, and new ones appear. A list that is short and current gets used; one that grows forever gets ignored. Keep only the words that went wrong in the last few clips and you will catch nearly everything that matters in a review of a minute or two.

04

Recording habits that help accuracy

Clear audio produces clear text. A microphone close to the mouth, one person speaking at a time, and a quiet room make a bigger difference than any setting afterwards. Cross-talk, where two people speak over each other, is the single hardest thing for any automatic system and for human typists too.

Spell unusual names out loud once near the start of a recording. It sounds formal, but it gives you a reference when checking the text, and it helps listeners who hear the name for the first time.

05

When typing by hand still makes sense

Content where every word carries legal or medical weight, such as a statement that must appear exactly as spoken, is worth checking word by word against the audio, whatever produced the first draft. The cost of one wrong word there is much higher than the time saved.

Heavily accented speech, songs and recordings with a lot of background noise also tend to need more hands-on work. In those cases, treat the automatic text as a rough draft and budget for a full read rather than a quick scan.

06

Correct the five words that matter

  • Let the ordinary words alone. They are almost always right and checking them is where the time goes.
  • Read only names, numbers and jargon. That is where the errors concentrate.
  • Correct them while the text is still editable. The text is editable up to that point.

Run a clip and read the names back.

Questions people ask

How accurate is automatic on clear speech?
High enough that corrections are a small handful per clip, provided the recording is clean. Poor audio is what changes that.
Which words go wrong most?
Proper nouns, technical vocabulary, numbers and anything spoken over another voice.
Is typing ever worth it?
For a single clip that matters enormously, occasionally. As a routine it is the thing that stops people posting.
Does timing suffer with automatic?
Timing follows the audio directly, so words appear as they are said. The variable is which word, not when.
What improves accuracy most?
The recording. A microphone close to the speaker does more than any correction afterwards.

Read next

01Add captions to videoGuide02Word by word captionsGuide03Burn captions into videoGuide