Turnwise / Qualitative research transcription

Transcription for qualitative research interviews

A practical guide to turning research interviews into checked, pseudonymised transcripts ready for coding, and how Turnwise handles each step on your own Mac.

The workflow, from recording to analysis

Most of the time in transcribing an interview goes into two jobs: getting words onto the page, and checking them against the audio. Turnwise does the first on your Mac and is built around the second. A typical study looks like this.

1. Record with transcription in mind

Speech recognition does best with clear audio. Put the recorder or microphone between you and the participant, avoid rooms with echo or background music, and do a short test before you start. For online interviews, record each side locally if your platform allows it. Turnwise reads MP3, M4A, WAV, MP4, MOV and most other audio and video formats, so you can use whatever your recorder or video call produces.

2. Transcribe on your Mac

Add recordings to a project and choose a quality setting: Standard uses Whisper large-v3-turbo, High uses Whisper large-v3. Both run locally through Apple's MLX framework, so the recording is never uploaded. You can set the language (Whisper covers 99) and add a glossary of names, places and technical terms that the model would otherwise mishear.

3. Speaker diarization: who said what

Turnwise separates the voices in the recording (speaker diarization, using the pyannote community-1 model) and splits the transcript into turns. You name each speaker once, for example Interviewer and P07. Diarization can go wrong when people talk over each other or sound alike, so the editor makes fixing it quick: press a number key to give a turn to a different speaker, or S to split a turn at the playhead.

4. Check the transcript against the audio

Automatic transcripts need checking before they are data. The editor is designed so you can do this from the keyboard while listening:

  • Tab or Space plays the current turn, rewinding a little so you hear the start.
  • Words the model was unsure of are highlighted, and ⌘G jumps to the next one.
  • S splits a turn, M merges it with the next, number keys reassign the speaker.
  • C marks a turn as checked, and a progress count shows how much of the interview you have verified.
  • Every change can be undone.

5. Pseudonymise

Record each participant's details and pseudonym in the project, then use find and replace to swap real names, places and employers for pseudonyms across the transcript. Keep the key linking pseudonyms to identities wherever your data management plan says it belongs, separately from the transcripts.

6. Code, or export to your analysis software

You can add codes and memos to turns in Turnwise as you read, which suits a first pass of familiarisation. For full analysis, export to NVivo, ATLAS.ti, MAXQDA or Dedoose, or to Word, plain text or CSV. See what each export contains.

7. Report your method

Turnwise records which model, version and settings produced each transcript and generates a methods statement from that record, so your write-up can say exactly how the transcripts were made and how they were checked.

Clean or verbatim?

How much detail a transcript needs depends on your method, and it is worth deciding before you start checking.

StyleWhat it keepsOften used for
CleanThe words as said, without fillers (um, er), false starts and repetitions.Thematic analysis, framework analysis, most applied and policy research.
VerbatimFillers, false starts, repetitions and timed pauses.Conversation analysis, discourse analysis, and any study where how something is said matters.

In Turnwise these are two views of the same transcript, not two transcripts. You check once and choose the style when you read or export. Verbatim view is a starting point for detailed notation; symbols for things like intonation or overlapping speech you add yourself.

Ethics, consent and transcription

Transcription is part of data processing, so it belongs in your ethics application, participant information sheet and data management plan. Questions reviewers commonly ask:

  • Who will hear or receive the recordings? If a transcription service or cloud tool is used, that is a third party handling identifiable data, and participants may need to be told. With Turnwise for Mac the recording is processed on the researcher's own computer.
  • Where are recordings and transcripts stored, and for how long? Turnwise stores its files on your Mac, so your institution's storage rules (encrypted disk, managed device, backups) apply as they would to any other file.
  • When are transcripts pseudonymised? Name the step and the tool. Find and replace in Turnwise lets you do it before anything is exported or shared.
  • How is accuracy assured? Describe the checking process: every turn is listened to and marked as checked, and the methods statement records the model and settings.

Requirements differ between institutions, funders and countries. Check with your ethics committee or data protection officer; this page describes the tool, not your obligations. More on this in private, offline transcription for research.

Transcribing interviews for a thesis or dissertation

Master's and PhD students often transcribe their own interviews, with little budget and an ethics approval that promised participants their recordings would stay with the researcher. Turnwise fits that situation: it runs on a recent Mac without an internet connection, keeps everything on your disk, and produces a methods statement you can adapt for your methodology chapter. Desktop licenses are currently free for students; write to [email protected] from your university address.

If your supervisor or department needs the transcripts in a particular format, the Word and plain-text exports work anywhere, and the analysis-software exports save a step if you are coding in NVivo, ATLAS.ti or MAXQDA.

Mac with Apple silicon (M1 or newer), macOS 14 or later. Desktop licenses are currently free for pilot users and students: write to [email protected].