How to Transcribe an Interview
A workflow for interviews where the quotes have to be right: recording, transcribing, verifying, and the consent question people skip.
The Scriptorium team ·
An interview transcript has one job that other transcripts do not: the quotes have to be exactly right, and you have to be able to prove it. That changes the workflow at both ends.
Before you record
Consent
Ask, out loud, on the recording. Recording laws vary: some places require every party to consent, some only one, and the rules differ again for phone calls and across borders. Getting agreement on tape at the start costs ten seconds and settles the question permanently.
Separately from the law, be clear about what the recording is for and whether the interview is on the record. That is a professional obligation, not a legal one.
The recording itself
This is where your transcript quality is decided:
- Put the microphone close to whoever is speaking. Closer than feels polite.
- Choose a room with soft furnishings. Avoid glass, bare tables and hard floors.
- Ask your subject not to talk over you and try not to talk over them. Overlapping speech is the single worst case for any transcription.
- Record each person separately if you can. Two tracks means no crosstalk problem.
- Do a ten-second test and listen back before you start the real thing.
Transcribing it
- Upload the recording to transcribe audio.
- Set the language.
- Rename the speaker labels to real names. Once each, and it applies throughout.
- Read the transcript with the audio playing.
Reading along is not optional for interview work. Automatic transcription produces fluent, confident sentences even when it has misheard, so a transcript that reads perfectly can still be wrong. The errors only reveal themselves against the audio.
Verifying quotes
For anything you are going to publish or attribute:
- Find the quote in the transcript.
- Click the line to jump the audio to that second.
- Listen to it. Check every word.
- Note the timestamp beside the quote in your notes.
The timestamp is the point. If the quote is challenged later, you can go straight to the moment rather than searching an hour of tape.
Check names, numbers, dates, job titles and organisation names specifically. These are where automatic transcription is least reliable and where an error does the most damage.
Clean or verbatim?
Decide before you start editing, because the two are different documents.
| Verbatim | Clean | |
|---|---|---|
| Includes | Every um, false start, repetition, stutter | The words, tidied |
| Use for | Research, legal work, speech analysis | Journalism, publishing, notes |
| Effort | Lower — leave it alone | Higher — you edit |
| Risk | Unflattering to read | Changing meaning while tidying |
If you clean, the rule is that you may remove disfluencies and you may not change what someone said. Cutting "I think maybe we should" down to "we should" is a different statement, not a tidier one.
Exporting
Timestamped text for your own working copy, so you keep the pointers back to the audio. Plain text for the version you write from. Markdown if you want a formatted document with speakers separated.
Keeping it
Keep the original audio as long as you might need to defend a quote. A transcript is a derived document; the recording is the evidence.