In this guide
Choose a transcription convention and an approved storage workflow before you begin. Label speakers consistently, mark uncertainty with timestamps, and review the completed transcript against the original recording before sharing.
To transcribe an interview, decide how much spoken detail to preserve, label each speaker, and turn the recording into text manually or with an approved transcription service. Add timestamps where they will help a reader find the audio, mark unclear words instead of guessing, and review the complete transcript against the original recording.
The right format depends on the purpose. A research team studying conversational behavior may need pauses and interruptions; an article draft may use an agreed readable transcript. Follow your project's consent, confidentiality, retention and editorial requirements before sharing or uploading the recording.
Choose a transcription style before you start
| Style | What it generally preserves | When it can help |
|---|---|---|
| Full verbatim | Words, repetitions, false starts and relevant non-speech events according to the chosen convention. | Research or review where the way something was said matters. |
| Clean verbatim | The spoken meaning with some agreed fillers or repeated starts removed. | Readable notes when the project permits those edits. Do not silently change meaning. |
| Edited summary | A condensed account rather than a word-for-word record. | Overview or internal notes; label it as a summary rather than a verbatim transcript. |
These labels do not define one universal standard. Agree on how to handle fillers, interruptions, laughter, timestamps and uncertain speech, then apply the rules consistently. KU Leuven's interview data guidance recommends a structured template and making changes such as anonymization clear.
Interview transcript example
The following is an invented example using speaker labels, timestamps and an uncertainty marker. The brackets are editorial annotations, not words spoken by the participant.
[00:00:00] Interviewer: What changed when the team moved to a quieter room?
[00:00:05] Participant 1: We could hear each other better, especially during the [unclear] meeting.
[00:00:12] Interviewer: Was that the weekly meeting?
[00:00:15] Participant 1: Yes, the weekly one.
Record the timestamp for an unclear phrase so a reviewer can return to it. Do not use the later question to silently invent the missing word. If the project requires strict verbatim, include the agreed pauses, false starts and overlap markings rather than assuming this example covers every convention.
1. Prepare the recording and permissions
- Confirm who may access the recording and transcript, what participants were told, and the approved retention and sharing rules. A voice can identify a person even if the filename does not.
- Keep the original recording and note its duration, source and any separate speaker tracks. Listen to the beginning, a section with overlapping voices and the end.
- Choose a convention for labels, timestamps, uncertain words, interruptions and redactions. Use project IDs or pseudonyms where required; keep any identity key separately with restricted access.
- Decide whether the material may be uploaded to an outside service. Consider the service's storage, access, deletion and organizational agreements, not just its accuracy.
Removing names alone does not guarantee anonymity: job titles, locations and distinctive experiences can identify someone. Erasmus University's qualitative-data guidance discusses planning anonymization and balancing privacy with research value.
2. Create the first draft
Manual: Play a short section, pause and type it, then repeat. Use consistent speaker labels and mark passages you cannot understand with a timestamp. Slower playback can help, but return to normal speed if the voice becomes distorted.
Automatic: If your organization permits it, use a service that supports the file, language and account. For example, Microsoft describes uploading supported audio in Word Transcribe; its documentation states that uploaded recordings are stored in OneDrive and that availability and usage limits apply. Speaker labels and recognized words need review.
The broader audio transcription guide compares Word, iPhone Voice Memos and manual workflows. Saturalabs does not generate transcripts.
3. Resolve speaker changes and difficult audio
- Play each speaker transition and confirm who is speaking. Automatic labels can switch or combine people, especially when voices overlap.
- Check names, numbers, dates, negatives and specialized terms against the original. Use project notes or ask an authorized participant when appropriate, but distinguish a later correction from what is audible.
- For overlapping or muffled speech, retain a timestamped uncertainty marker when you cannot verify the words. A plausible guess is not a verified quote.
- If a processed version helps you hear a passage, compare it with the original. A cleaned or isolated voice may omit or alter sounds and is not proof of missing words.
For a supported short MP3 or WAV, the local audio trimmer can prepare an excerpt for repeated listening. Record the excerpt's original start time so its timestamps can be mapped back. It creates an audio file, not a transcript; consult its duration and file-size limits.
4. Review and deliver
- Listen through the entire interview against the text; check that no answer or question was accidentally omitted.
- Apply the agreed style consistently and make redactions or substitutions clear according to the project convention.
- Review names and contextual details for identification risk before distributing a research transcript.
- Store the original, working draft, identity key and final version according to the approved access and retention plan.
- When appropriate, ask the interviewer or an authorized reviewer to check uncertain passages before publication or analysis.
The Trinity College research guidance provides examples of masking identifiers when anonymity has been promised. The requirements for a particular project depend on its own agreements and policies.
Common questions
Do I need timestamps on every line?
Not always. Agree on the frequency and format with the people using the transcript. Speaker changes and uncertain passages are useful places to consider timestamps.
Should I remove every “um”?
Only if the agreed transcription style allows it. Fillers, pauses and repetitions can be meaningful for some research, so document the chosen rule.
Can an automatic transcript identify speakers correctly?
It may propose speaker labels, but verify them against the audio. Similar voices, interruptions and overlapping speech can produce mistakes.
Can Saturalabs transcribe an interview?
No. Its audio tools can prepare supported excerpts for a separate workflow; they do not produce the transcript.