Interview transcription with speaker labels: a quick guide
Interview transcription with speaker labels: how to record so voices stay separate, get a draft that shows who said what, and check it in ten minutes.
You have an hour of recorded conversation and you need text that shows who asked and who answered. Journalists, researchers, recruiters and students do this all the time. It used to cost an evening of typing. Today a draft arrives in minutes, but its quality depends heavily on how you recorded. Here is a workflow that gives a clean transcript with speaker labels.
What speaker labels actually are
The technical name is speaker diarization. The software works out who spoke when, and separately recognises what was said. The result reads "Interviewer: ..." and "Guest: ...", not one wall of words. For an interview this matters: a quote without a speaker is useless.
One caveat. Diarization is not perfect. Vendors admit accuracy drops with overlapping speech, similar voices or poor audio (Sonix's overview of diarization), and research confirms that overlapping speech causes more speaker confusion (paper on arXiv). So plan a short manual check after the software finishes.
Step 1. Record so the voices are easy to tell apart
The recording matters more than the tool.
- A quiet room and a microphone close to each person. A phone lying between two people is worse than two headsets.
- Avoid talking over each other. A one-second pause after an answer saves minutes of fixing.
- Start with introductions so every voice has a name you can match to a label.
- For an interview over a call, record from your own computer so both voices land in the file. Our guide to recording and transcribing calls on Mac covers Zoom, Teams and Meet.
Tell the other person you are recording and get their consent. For journalism this is also a professional norm.
Step 2. Get a draft with labels
You have three options: upload a file to an online service, use a file-transcription app, or record the conversation directly in an app that separates speakers itself.
Tiho is the last kind. On iPhone, interview mode records a meeting and separates who said what, and you can rename the speakers. On Mac, calls are recorded in full with sound and screen, and Tiho Pro adds speaker names, filtering and export. Transcription works in Russian, English, Polish, Spanish and more, so a mixed-language interview does not need splitting into two files. The original audio stays next to the text, so you can replay any doubtful line. See the iPhone section of the home page for details.
Honest limits: Tiho is not a certified court-grade transcript, and the iPhone app is currently in closed testing. Free recordings run up to 30 minutes, paid sessions up to 60, and Pro gives more room for long conversations. Check the pricing page for current terms.
Step 3. Check in ten minutes, not two hours
Do not read everything line by line. Go to the risky places.
- Speaker labels in the first minutes. If they are right at the start, they are usually right later. Rename speakers once.
- Interruptions and laughter. Labels fail here most often. Find them and listen again.
- Quotes you will publish. Check each against the audio word for word.
- Names, titles, numbers. Verify spelling and figures. If a word keeps failing, fix it once and add it to the dictionary.
- Off-topic chatter. Decide what stays in the piece.
Step 4. Turn the transcript into something usable
A raw transcript is not a finished text. Typically you:
- mark key quotes with their timestamps;
- trim filler where speech is paraphrased;
- write a short summary: main points, agreements, what to verify;
- dictate your own notes right after the conversation. Voice typing for students describes the same habit, and it works equally well for interviews.
FAQ
How accurate is speaker separation?
In good conditions, with clean audio and people taking turns, it usually works well. Vendors cite roughly 80 to 95 percent, but those are their own figures, not independent tests. With interruptions and noise there are more mistakes, so always review.
Can I transcribe an interview that mixes two languages?
Yes, if the app recognises several languages in one recording. Tiho understands Russian, English, Polish and others, including when they alternate. Still check the spelling of rare terms.
Do I need the other person's consent?
Recording laws depend on the country and situation, and this is not legal advice. The safe practice is to say at the start that you are recording and to get consent, ideally on the recording itself.
Does it work offline?
Tiho dictation can run on your device without internet. For confidential interviews, read the privacy section and choose the on-device option where it is available.