How to transcribe an interview: a practical guide

A transcript is only as good as the recording behind it and the checking after it. This guide walks through the whole path — before, during and after — so the quote you print is the quote that was said.

Before you record

Most transcription problems start at the microphone. A few minutes of preparation save an hour of guessing later.

• Get close to the speakers. A phone on the table between two people works; a phone in a pocket or across the room does not. Distance hurts more than a cheap microphone does.
• Choose a quiet room. Background music, a coffee machine or traffic compete with voices, and people talking over each other are hard to separate for any listener, human or machine.
• Record a short test. Thirty seconds, played back on headphones, tells you whether both voices are clear.
• Ask for permission. Tell the other person you are recording and what the recording will be used for, and get their agreement. The rules differ between countries and situations, so check what applies to you.
• Write down names and terms. Names of people, companies and places, and any jargon, are the words automatic transcription gets wrong most often. A list made before the interview helps later.

Turn the recording into text

Typing an interview out by hand usually takes several times longer than the recording itself. Automatic transcription does the first draft in minutes; your time goes into checking it.

1. Upload the recording. Common audio and video files such as MP3, M4A, WAV or MP4 work as they are — there is nothing to convert first.
2. If you know the language and the number of speakers, set them. It helps the engine separate the voices correctly.
3. If the interview is full of names or technical terms, add them to a glossary first, so they come back spelled right.

Check before you quote

A machine transcript is a draft. Before any sentence leaves your notes, check it against the recording.

• Listen to every passage you intend to quote. In SonaFact every line carries its moment in the recording — to the word in 32 languages — so you can click its time and hear exactly what was said.
• Look at the words the engine was unsure about. SonaFact underlines them; they are the first place to check.
• Watch the small words. A missing "not", a wrong number or a misheard name changes the meaning more than any typo.
• Check who said it. Speaker separation is automatic and usually right, but when two people talk at once a line can land with the wrong person. Move it to the right speaker before you quote it.

Name the speakers and decide how clean the text should be

Replace "Speaker 1" and "Speaker 2" with names or with codes if the interview is confidential. In SonaFact you rename a speaker once and the change applies to the whole transcript.

Then decide what kind of transcript you need:

• Verbatim keeps everything as it was said, including repetitions and false starts. It is what you want for research or when the exact wording matters.
• Edited removes hesitations and repetitions so the text reads well. It is fine for notes and articles, as long as quotes keep the speaker's meaning. When you shorten a quote, mark the cut with an ellipsis.

Export and keep the original

Export the transcript in the format you work with — DOCX or PDF for documents, Markdown for notes. Keep the original recording for as long as you may need to prove what was said, and delete both when you no longer need them. If the interview is confidential, check where your transcription service stores recordings and what happens when you delete them — our checklist lists the questions worth asking.

Checklist

• Recorded close to the speakers, in a quiet room, after a short test.
• Permission to record, and to transcribe, agreed.
• Names and terms written down or added to the glossary.
• Every quoted passage checked against the recording.
• Speakers named, and quotes attributed to the right person.
• Original recording kept for as long as needed, then deleted.

Try it on your own recording

The free plan gives you 3 transcripts a day, up to 30 minutes each, without a card — enough to judge the transcript, the speakers and the editor yourself.