Interview transcription on a Mac: formats, consent, examples

Before you record: consent and what you promise

Ask before you press record, and say what will happen to the recording: who hears it, whether it's uploaded anywhere, how long you keep it, and whether quotes will be attributed. Research ethics boards and editors expect this, and the law in many places does too. Our guide to whether it's legal to record a meeting covers the consent rules by country.

Whatever you promise constrains the tool you can use afterwards. If you told a participant the recording stays with you, uploading it to a transcription service breaks that promise, however convenient.

Three ways to get the transcript on a Mac

• macOS itself: if you recorded the interview in Voice Memos, a Mac with macOS Sequoia and Apple silicon can transcribe it; open the recording, click the Transcription button, then select, Control-click and Copy. Notes does the same for audio recorded inside a note on Macs with M1 or later in a listed set of languages. Neither imports a recording made elsewhere, and neither labels speakers.
• A hosted service: Otter's pages say to log in or create a free account and upload the file. The free plan is listed at 300 minutes a month, files up to 90 minutes, and 3 pre-recorded file imports per user, lifetime; Pro ($16.99 per user per month, $8.33 billed yearly) allows 10 imports a month. Languages are English, Spanish and French. The file is uploaded and the transcript stays in the account, which is the point to weigh against your consent form.
• On the Mac: QuietNote is a note taker for Mac that imports a common audio or video file and transcribes it on your Mac with the Whisper model that ships with the app, offline. The output is a Markdown transcript with timestamps in a folder you choose; no account, nothing uploaded. A summary is optional, with a local model through Ollama or Claude with your own API key (transcript text only, only when you ask). The first signed Mac release isn't out yet; you can leave your email for it. It needs macOS 14.2 or later.

Transcript formats: clean verbatim and intelligent verbatim

Pick the style before you start editing, because it decides how much of the raw transcript you keep. Conversation analysts need every hesitation; a journalist needs what was meant. The two common styles:

• Clean verbatim (sometimes called full or true verbatim): every word as spoken, including false starts, repetitions, fillers and non-verbal sounds, with speaker labels and timestamps. Used in qualitative research where how something was said matters.
• Intelligent verbatim: the same content with fillers, false starts and stutters removed and grammar lightly tidied, without changing meaning. Used for journalism, oral history summaries and most business interviews.

A speech model gives you something between the two: most fillers dropped, no laughter markers, no speaker labels. Turning it into clean verbatim means listening again; turning it into intelligent verbatim is mostly labelling and light editing. For the note-taking side, the interview notes template gives the transcript a structure for themes, quotes and follow-ups.

Speaker labels on a single recording

Most interviews are recorded on one device in the room or one side of a call, which means one mixed track. A speech model transcribes words, not identities, so it can't reliably say who spoke. Label speakers yourself using the timestamps; with two people and one interviewer it's quick, since the questions give the structure away.

If you can, record each person on their own track and transcribe the tracks separately: each file is then one voice. A stereo recording with one person per channel can be split into two mono files for the same effect. For a call recorded on the Mac, QuietNote separates your microphone from the other side's audio, which gives you two labelled voices for a one-to-one interview.

Anonymising and keeping the recording private

• Replace names with codes: P1, P2 or a pseudonym, consistently, in the transcript before it goes anywhere, including before any cloud summary step.
• Remove identifying detail: employers, places, dates and relationships that would identify someone even without a name.
• Keep the raw recording and the anonymised transcript apart: the recording in one encrypted folder, the working transcript in another, and a key file linking them only if your protocol allows it.
• Prefer tools that don't upload: transcribing on the Mac means the only copy of the audio is the one you control. If you do use a cloud summary, send the anonymised transcript, not the raw one.
• Delete on schedule: note the retention period you promised and act on it.

Quality points that matter for interviews

• Microphone distance: a phone on the table between two people picks up the room. Closer is better, and a quiet room matters more than any setting afterwards.
• Accents, jargon and names: these cause most errors. Keep a list of correct spellings and fix them in a pass.
• Overlap: people interrupt each other; on one track those lines come out garbled. Ask for pauses if the protocol allows.
• Accuracy: no tool can promise a percentage for your recording. Apple says of its own Live Captions that accuracy may vary; Whisper's README notes its English-only model variants tend to do better for English. Listen to anything you'll quote.
• Time: transcription usually takes about the length of the recording on a hosted service (Otter's page gives a five-minute file taking about five minutes); on a Mac it depends on the Mac and the model size. The editing to a verbatim standard takes longer than either.

What to do with the transcript

• Coding and themes: a plain-text transcript imports into any qualitative analysis tool and is searchable in an editor.
• Quotes: pick them with timestamps so you can re-check wording against the audio before publishing.
• Summary: a one-paragraph summary per interview, written or generated locally, helps when you have twenty of them.
• Follow-ups: note questions you didn't get to and promises you made to the participant.