Skip to content
Sonorarium
Back to blog

Transcribing Interviews and Focus Groups for Qualitative Research

Verbatim or clean, conventions, anonymizing and pseudonymizing, consent and GDPR, timestamped quotes, importing into NVivo, MAXQDA or ATLAS.ti, and focus group voices.

Sonorarium Team 5 min read

Short answer: before you transcribe, settle your transcription convention (verbatim or clean) and how you will pseudonymize. Transcribe each recording separately, review while listening, and replace identifying details. Then export to Word: every paragraph starts with a [hh:mm:ss] stamp, a pattern that the MAXQDA and ATLAS.ti documentation lists as importable. In a focus group, Sonorarium does not separate speakers, so you label each participant yourself while reviewing.

If you have just a few interviews for a final project, the basic workflow is in how to transcribe interviews for your dissertation or thesis. This guide is for when the analysis is the heart of the project: a corpus of many interviews that you will code in NVivo, MAXQDA or ATLAS.ti, focus groups, or a transcription and pseudonymization system shared across a whole project or team.

1. Pick the convention before the first interview

Switching conventions mid-project means redoing work. The usual options:

What you analyze Typical convention What to add by hand
Content: themes, codes Clean (intelligent) verbatim Little: long pauses, laughter
Conversation structure and intonation Jefferson (2004) or GAT 2 (Selting et al., 2009) Pauses, overlaps, intonation: everything
Discourse with fillers and repetitions Strict verbatim Relevant fillers and hesitations

Automatic transcription lands close to clean verbatim. Treat it as a draft and add whatever your convention demands. The differences, with examples, are in verbatim vs. clean transcription. If your lab or program has its own convention, follow that.

2. Consent and data protection

This isn't legal advice; it's a list of questions for your ethics board (IRB or REC) or data protection officer before you record.

  • The consent form should cover recording, transcription with an online tool, where data is stored and for how long, pseudonymization and withdrawal. For interviews, the French regulator CNIL recommends separating consent to data processing from consent to take part in the research (CNIL, in French).
  • Pseudonymizing is not anonymizing. Under the GDPR, pseudonymized data (a name replaced by a code, the key kept separately) is still personal data; anonymous data falls outside it (Art. 4(5), Recital 26). For research, Art. 89 lists pseudonymization among possible safeguards where the purpose can still be met. Other countries have different regimes, so check yours.
  • What Sonorarium does with your files. Audio is deleted automatically after 7 days, the transcript in your account stays until you delete it, and nothing is used to train models. The privacy policy lists the processors and transfers outside the European Economic Area. If your institution requires a data processing agreement, sort that out before you upload anything.
  • Names come in with the audio. If a participant says their name, it appears in the first transcript: pseudonymize afterwards, in the text.

3. Transcribe and review

  1. Upload each recording as its own file, named with the participant code (up to 2 GB and 5 hours).
  2. Choose the language and press Transcribe.
  3. In the editor, click a passage's timestamp to hear it and correct the text. Proofread within the 7 days after uploading: after that the audio is deleted and you can no longer listen to it.

The first 15 minutes are free when you create an account, with no card. After that you pay per minute, with no subscription and no expiry: 30 min for $2, 120 for $6, 300 for $14, 900 for $39 and 3600 for $139. Twenty one-hour interviews are 1,200 minutes: for example one 900-minute pack and one 300-minute pack ($53). See pricing.

Check proper names, acronyms, discipline-specific terms and passages with crosstalk. Downloading and editing need your account.

4. Pseudonymize once, consistently

  • Assign codes (P01, P02…) and keep the key (real name and code) apart from the transcripts.
  • Use the editor's search box to find every mention of a name and replace it the same way every time: "[City A]", "[Hospital 1]".
  • Watch for indirect identifiers: a rare job title plus a small town identifies anyone.
  • Export only afterwards: the Word, TXT and SRT files contain the corrected text.

5. Quoting with timestamps

In the downloaded Word file, every paragraph begins with [hh:mm:ss] (the TXT has no timestamps). A reference looks like this: "…we had never looked at it that way" (P03, 00:41:12). The stamp marks the start of the passage, not the word: it takes you back to the audio but isn't an exact measurement. If your department wants line or page numbers, put them next to the time.

6. Exporting to NVivo, MAXQDA or ATLAS.ti

This is what each vendor's documentation says; we haven't tested your version, so import one trial interview first.

  • NVivo. Imports Word and TXT; in Word it accepts paragraphs that start with a valid timestamp (hh:mm:ss) or tables with a timestamp column. The help page doesn't mention square brackets: if they aren't recognized, remove them with Find and Replace (NVivo).
  • MAXQDA. Import tab → Transcripts → "Transcript with Timestamps" (or your software's entry). [hh:mm:ss] is among the recognized timestamp patterns. If you select the audio file during import, MAXQDA links it to the text; the timestamps move to a column and a click plays that point (MAXQDA).
  • ATLAS.ti. Accepts VTT, SRT, TXT, DOCX and RTF. For TXT, RTF and DOCX the documentation lists [hh:mm:ss] among supported timestamps (ATLAS.ti). Sonorarium's SRT or VTT is another route.

Sonorarium's Word file starts with a heading (the file name). If your software imports it as text, delete it.

7. Focus groups: several voices, no automatic separation

Sonorarium doesn't tell who is speaking. A workflow that works:

  1. While recording: ask everyone to say their name or number at the start, and have the moderator address people by it. One person at a time: overlap is what transcribes worst.
  2. With an observer: a second person notes the order of contributions.
  3. While reviewing: put "M:" (moderator), "P1:", "P2:"… at the start of every turn as you listen. Clicking the timestamp helps confirm the voice.
  4. Mark overlaps with a fixed tag, for example [overlap].
  5. In your analysis software, these labels let you code by participant.

Expect to spend more review time on a focus group than on a one-to-one interview.

FAQ

Is automatic transcription acceptable for research data? Yes, if you review the result, report it in your methods and your ethics board allows it.

What about dialects or less common languages? More than 60 languages are supported; choose the dominant one. With a strong dialect, expect more review.

How do I identify each participant in a focus group? Sonorarium doesn't separate speakers, so you label them by hand while reviewing, using the timestamps; the method is in how to transcribe audio with multiple speakers.

More in the research transcription and focus group transcription guides.