Interview transcription: method, template and GDPR

Verbatim or clean: choose before you start

• Full verbatim: everything that is said, hesitations, repetitions, laughter, silences and prompts included. For discourse or conversation analysis.
• Clean transcript: all the meaning, without the “ums”, false starts or verbal tics. For thematic analysis, reporting or quotes.
• Summary: a structured summary with a few exact quotes. To keep a record of a secondary interview, not to analyze it.

By hand, four to six hours of work per hour of recording is a common estimate. An automatic first draft, reviewed afterwards, cuts that time sharply, especially for a clean transcript.

Set your conventions

Write your rules once, at the start of the project, and stick to them from one interview to the next: that is what makes transcripts comparable.

• Speakers: a short, stable code (INT for the interviewer, P07 for the participant), never the real name.
• Timestamps: [mm:ss] at the start of each turn, or every minute.
• Uncertain passages: [inaudible 12:40], [unsure: Bertoua?].
• Non-verbal: [laughs], [silence 5 s], [interruption].
• Removals: [name removed], [place removed], for anything identifying.

A template to copy

• [00:00] INT: Are you okay with me recording? Your name won’t appear anywhere.
• [00:06] P07: Yes, fine.
• [00:09] INT: How long have you worked at the market?
• [00:14] P07: Since 2019. Before that I was in [place removed], with my uncle [name removed]. [laughs]
• [00:31] P07: The checks started after [inaudible 00:33] the year after.

Step by step

1. Record cleanly

   Consent stated and recorded at the start, a quiet place, the phone close to the person. Good audio saves hours of review.

2. Make an automatic first draft

   A speech recognition model such as Whisper produces a timestamped text in minutes. On your own device, the audio goes nowhere.

3. Review while listening

   Fix proper names, numbers and mixed-language passages. Listen again to every sentence you will quote word for word.

4. Pseudonymize

   Replace names, precise places and identifying details with codes; keep the correspondence in a separate table, stored elsewhere.

5. Archive and delete

   Decide how long you keep the audio. Often, once the transcript is reviewed, the recording is no longer needed.

The GDPR without jargon

• The voice and content of an interview can identify a person: they are personal data.
• Pseudonymized data (names replaced by codes, key table kept apart) are still personal data: the GDPR still applies.
• Only effective anonymization, which makes re-identification impossible in practice, takes data out of the GDPR. A detailed interview rarely gets there.
• If you send the audio to an online transcription service, that service processes personal data on your behalf: it is a processor under Article 28, and you need a contract with it (often called a DPA).
• Transcribing on your own device avoids that transfer: no third party, no processing contract for that step.

With nTilia

Record the interview into the note, or import an audio file. From Pro, nTilia transcribes it with Whisper on the device, with no network: the transcript lands in the note with [mm:ss] timestamps. You review it, replace names with codes, then the AI can, on request, suggest a cleaned-up version, which you compare Before / After before accepting it.

The whole notebook is encrypted at rest by the vault (AES-256-GCM, on the free plan). An interview in a “device” note never leaves the phone. And if the interview feeds an article, the journal publishes it with a Cite block and, from Pro, a Zenodo DOI.