How to transcribe an interview recording
Get the interview as a file, then pick one of three ways. One-off interview: Nerpa's free web tool - drop the file at nerpa.work/tools/free-transcription, no signup, up to 35 minutes, text with speaker labels in minutes. Interviews regularly: the Nerpa Mac app - on-device transcription with speaker labels (on a supported Mac), so the interviewer and subject stay separated. Never upload anything: local Whisper - free but fiddly. The three ways compared, plus the editing pass that turns a raw transcript into a publishable document.
Step 1: get the recording as a file
Every source produces something a transcriber can read - the formats rarely matter anymore:
- Phone or voice memo: export as .m4a (on Mac, right-click the memo and Export - the full voice memo storage and export walkthrough is in a separate guide).
- Zoom, Meet, or Teams recording: download from the platform; m4a or mp4 both transcribe fine.
- Recorded on your Mac with Nerpa: the file is already in a local folder - the transcription pipeline picks it up automatically.
- Video interview: no need to extract audio first; the same file transcribes with the video track ignored.
One quality note before you hit record: a close microphone on the subject beats any transcription software. Background noise and distant audio are the main causes of garbled words - no tool fully rescues a bad recording, though the same rules for clean capture apply to interviews as to webinars.
Step 2: pick a transcription method
The right method depends on volume and sensitivity, not on the file:
- Free web tool - for the one-off. Drop the file, wait a few minutes, get text with timestamps and speaker labels. No account. Limits are honest: 35 minutes per file, three files a day, and the file is deleted right after transcription. Start here if you've never transcribed before.
- Nerpa Mac app - for the working journalist or researcher. Transcription runs on your Mac (on a supported Mac) - no per-minute metering. Speaker labels separate the two voices, and every interview is filed under the subject's name, so an archive builds itself. Raw audio stays in a local folder on your Mac.
- Local Whisper - for sensitive sources. whisper.cpp or a wrapper app transcribes entirely on your machine; nothing is uploaded anywhere. The cost is setup effort and slower turnaround on older Macs - the full trade-offs are in the voice memo transcription guide, which covers the same three-way split.
Step 3: verify the speaker labels
An interview is a two-voice document, and speaker labeling (diarization) is what makes it readable - "Q:" and "A:" instead of a wall of text. Machine labels are good but not perfect:
- Read the first ten exchanges and confirm the interviewer and subject are consistently labeled.
- Check the turn boundaries - mislabels cluster where speakers interrupt or talk over each other. The full label-fixing routine: transcribing with speaker labels.
- Fix a wrong label with one click in the editor; catching a systematic swap early saves a full re-read.
If your tool produced an unlabeled wall of text, the fastest manual fix is a two-column pass in a text editor: keep the subject's answers, delete or compress your own questions you already know.
Step 4: the editing pass
The transcript out of the machine is accurate but not yet readable. A publishable interview needs one deliberate pass:
- Decide on verbatim vs readable before editing. Verbatim keeps every "um" and false start (legal, academic); readable trims fillers and stumbles while keeping the subject's words (journalism, blogs). Mixing the two styles mid-article is what editors catch.
- Never paraphrase the subject's answers. Trim their words, reorder for flow if needed - but a quote that changed the wording isn't a quote.
- Quote-check against audio. Any line you'll publish as a direct quote, replay from the timestamps in the transcript. This takes seconds per quote and is the difference between confidence and correction notices.
- Budget the time honestly: the transcription itself is minutes; the editing pass is 30-60 minutes per hour of audio. Planning around the edit, not the transcription, is what keeps deadlines.
Interviews that transcribe themselves
Nerpa records the interview without a bot joining, transcribes on your Mac with speaker labels, and files every conversation under the subject - the archive builds itself while raw audio stays in a local folder.
Latest stable release. Requires macOS 12+
Three ways compared
| Free web tool | Nerpa Mac app | Local Whisper | |
|---|---|---|---|
| Setup | None - open the page | Install once, sign in | Real setup - models, CLI or wrapper |
| Speed | Minutes per file | Minutes, unlimited files | Depends on your hardware |
| Speaker labels | Yes | Yes, plus editing | No - DIY |
| Limits | 35 min/file, 3 files/day | Unlimited transcription | None |
| Where audio goes | Deleted right after transcription | Stays in a local folder on your Mac | Never leaves your machine |
| Best for | One-off interviews | Regular interview work | Sensitive sources |
The short version: volume pushes you from the web tool to the Mac app; sensitivity pushes you to local Whisper; most people start with the first and land in the second.
Frequently asked questions
What is the best way to transcribe an interview?
For a one-off interview, a free web tool is fastest: drop the file, get text in minutes, no signup. For journalists and researchers doing regular interviews, a Mac app like Nerpa is better - on-device transcription on a supported Mac, speaker labels that separate the interviewer from the subject, and quotes you can verify against timestamps. Fully local Whisper is the choice when the recording must never leave your machine.
How long does it take to transcribe an hour-long interview?
Machine transcription of an hour-long interview typically takes a few minutes of processing - far faster than real time. The editing pass is what takes time: budget roughly 30-60 minutes to clean up filler words, verify speaker labels, and check quotes you plan to publish. Manual transcription by comparison runs 4-6 hours per hour of audio.
How do I keep two speakers separate in a transcript?
Use a tool with speaker labeling (diarization): it splits the transcript by voice and labels each turn. Nerpa labels the interviewer and subject automatically; the free web tool on this site also returns speaker-labeled text. If labels get mixed up at a turn boundary, fixing one label in the editor is faster than re-recording.
Is it legal to record and transcribe an interview?
Recording laws vary by jurisdiction: some places require one-party consent, others require all parties to agree. The safe default is to announce the recording at the start and get verbal consent - it protects the publication and rarely bothers a subject. For sensitive sources, store the recording locally and check your editor's or institution's policy first.