How a Spoken Interview Becomes Clean Text, Step by Step

The first pass on any recording involves no typing at all. I listen through the whole file at normal speed with a notepad beside me, marking where the speaker changes pace, where the room noise rises, and where a name or a technical word shows up. This pass takes about as long as the recording itself, and learners often want to skip it. I ask them not to, because it tells me how the real work will go.

A calm interview in a quiet office needs a very different plan from a panel recorded in a hall with a clattering coffee station. By the end of the listen I know roughly how many passes I will need, which stretches will be slow, and whether I should ask the person who sent the file for a glossary or a speaker list.

Skipping this step usually costs more time than it saves. A word misheard in minute two and corrected in minute forty means searching the whole draft for every place it was wrong.

• Listen once at normal speed without typing
• Mark names, acronyms and unclear stretches
• Note room noise and pace changes
• Decide how many passes the file needs

The second pass is the draft. I type what I hear, in order, and I do not stop to polish. When a stretch is unclear I insert a short marker with the time, such as a bracketed query, and keep moving. The goal is a complete skeleton of the speech, not a finished text.

Working this way protects the rhythm of listening. If I stop at every doubtful word, I lose the thread of the sentence and then mishear the next one too. A marker costs a second to write and gives me a clean list of problems to return to.

In class we practise on short sample recordings, and I ask learners to count their markers afterwards. A first draft with a dozen markers in five minutes of speech is normal. It means the learner was honest about what they could not hear, which is the habit that makes later review quick.

Terminology is where drafts quietly go wrong. A speaker says a product term, a place name or a surname, and a plausible wrong word gets typed because it sounds close. So after the draft I build a short list of every proper name and specialist term and check each one against whatever the client supplied: a glossary, a programme, a slide deck, or an earlier transcript.

When nothing was supplied, I ask. A single message asking how a guest spells their name saves a correction later and shows care for the person being quoted. If no answer comes, I pick the most likely spelling, keep it identical throughout, and flag it so the reviewer can confirm it.

Consistency matters as much as correctness here. A name spelled two ways in one transcript looks careless even when one of them is right.

• Check every name against supplied material
• Ask rather than guess when a spelling is open
• Use one spelling throughout and flag doubtful ones

Unclear audio is not a single problem. Some stretches are quiet, some are covered by another voice, some are muffled by distance from the microphone. Each one gets a different treatment. For quiet speech I raise the playback volume slightly and slow the speed. For muffled speech I replay it in short loops of two or three seconds, often a dozen times, until the shape of the words settles.

Context helps more than sharper ears. What was the question just before? Which terms has this speaker used already? Often the sentence completes itself once I stop staring at the one word I cannot hear.

Some stretches stay unclear no matter how many times I replay them. In that case I mark them honestly with a short note instead of inventing text. A readable gap is better than a confident error that misleads the viewer.

• Slow the speed for fast or quiet speech
• Replay short loops for muffled words
• Use the question and earlier terms as clues
• Mark what truly cannot be heard

Once the draft is complete and the markers are resolved, I clean it. Cleaning means removing false starts, repeated filler words and stumbles that would only distract a reader, while leaving the speaker's meaning and manner untouched. A person who speaks in short, blunt sentences should still read that way afterwards.

This is a judgment call, and I explain it to learners as a matter of purpose. Captions for an event recording usually benefit from light cleaning so they stay readable at speed. A transcript kept for an archive may need to stay closer to the spoken words. The client's wishes decide where the line sits.

I also fix punctuation at this stage, because commas and full stops steer how a later caption line will break. Good punctuation now makes the timing work in the next module noticeably easier.

The last step is a read-through with the audio playing, at normal speed, eyes on the text. This catches the things a silent reread misses: a dropped word, a sentence that ends too early, a number typed wrongly. I read it aloud under my breath when something feels off.

Only then does the text move on to timing. Handing over a clean transcript means the next stage starts from solid ground, and the person doing the timing is not forced to guess what a speaker meant.

On a short interview this whole routine takes a sensible multiple of the recording length, and the first few attempts take longer for every learner. That is expected. Speed comes from the habits above, not from rushing the listening.

• Read with the audio playing, at normal speed
• Check numbers, names and sentence endings
• Hand over a transcript that needs no guessing