How to Transcribe an Interview

Published Nov 10, 2026

An interview transcript lets you quote precisely, search what was actually said, and stop scrubbing through audio every time you need one sentence. Getting there is a three-stage job: prepare the recording, convert speech to text, then clean up the result. Skipping the first or last stage is what makes transcription painful.

This guide walks through all three stages for the two common cases — a quick transcript for your own notes, and an accurate record you can publish or cite — and is honest about where automatic transcription still needs a human.

Related platform workflows

For public video links, choose TikTok, Instagram or Facebook. Caption availability and timed exports differ by platform.

Before You Start: Audio Quality and Release Forms

Transcription quality is decided before any software touches the file. A recording where both speakers sit close to the microphone, in a quiet room, transcribes dramatically better than a café capture on a phone across the table. If you can still influence the recording: use one mic per speaker when possible, ask participants to avoid talking over each other, and note the spelling of unusual names while you can still ask.

The other pre-flight item is permission. Recording people and publishing what they said can involve consent and data-protection rules, and the requirements differ between journalism, research, and workplace interviews. Check the applicable consent and recording rules in your location, and get release forms signed before the interview rather than after — this guide is practical advice, not a legal conclusion.

Also decide your target format now: verbatim (every "um" included), clean verbatim (speech smoothed slightly), or a summary with key quotes. The choice changes how long cleanup takes, and it is easier to decide before you have text in front of you.

Method 1: Do It Manually and When That's Actually Fine

Manual transcription has a reputation as the punishment option, but for short or difficult material it is often the right one. If you need two minutes of quotes from a ten-minute interview, typing those two minutes beats correcting forty minutes of machine errors. Accents, industry jargon, crosstalk, and poor audio also degrade automatic results enough that typing from scratch can be faster.

A manual session works best with a few mechanical habits:

Open the audio in a player that supports speed control and a rewind shortcut, and your transcript in a window beside it.
Work in short passes — play a sentence or two, pause, type, repeat — rather than trying to type along live.
Drop a timestamp every few minutes so you can jump back to the source audio later without hunting.
Mark unclear words with a placeholder like [inaudible 00:14:30] instead of guessing, then resolve them in one pass at the end.
Label speakers consistently from the start (S1/S2 or initials), even if you fix the labels later.

Manual work is slow — expect the total time to be a multiple of the audio length, growing with the number of speakers. That cost is the price of exact control over every name and number.

Method 2: Use Automatic Transcription

For anything longer than a few minutes, automatic transcription does the heavy lifting. Your options, roughly in increasing order of cost and accuracy:

Speech-to-text apps and services that accept an audio or video file and return text in minutes. Quality varies by service, language, and audio.
Human transcription services, where people type and check the result. Slower and paid, but far fewer errors — the standard when the transcript is going public.
A hybrid: run automatic transcription first, then edit its output instead of typing from zero. For most interviewers this is the sweet spot.

Audio to Text includes automatic speaker labels. One thing automatic transcription does not do well on its own is telling speakers apart. Some services attempt speaker diarization (splitting text by who spoke), but results still need review whenever two people talk at once. Whatever produced the first draft, the cleanup pass is yours.

For a local recording, use Audio to Text: sign in, upload MP3, M4A, WAV or MP4 up to 4 MB, and process its first 10 minutes with speaker labels. Rename or merge labels in downloaded TXT, not in the current interface. For a public YouTube interview with captions, use the YouTube transcript tool and read or download its text — the caption-reading caveats from that platform apply.

Cleaning Up Speaker Names and Numbers

Every draft transcript — typed or automatic — needs a cleanup pass, and most of the errors hide in the same places:

Names. Automatic services mangle unusual spellings; do a find-and-replace once you know the correct form.
Numbers. "Nineteen ninety-five" might be a year or a quantity; verify against the audio whenever the number matters.
Technical terms and product names. The software has never heard of your niche vocabulary; you have.
Sentence boundaries. Spoken pauses become periods at random; read the text aloud to find where sentences actually ended.

Work through the text once for speaker labels, once for names and numbers, and once reading against the audio for anything you plan to quote. Three fast passes beat one exhausted one.

How to Store and Search Your Transcripts

A transcript is only useful later if you can find the right one and the right line in it:

Keep plain text as the master format — it opens everywhere and searches reliably. Other formats can be exports.
Name files by date and interviewee, and keep the audio file next to the transcript so quotes can be verified.
Leave your timestamp placeholders in place even after cleanup; they are what lets you jump back to source audio years later.
If interviews accumulate, a single folder per project with a simple naming scheme beats almost any database until you have dozens of transcripts.

Finally, apply the same caution you would to the audio itself: an interview transcript can contain personal data, so store and share it with the same care you promised when you asked for the interview.

Frequently Asked Questions

How long does it take to transcribe an interview manually?

Expect the work to take a multiple of the audio length — several hours for a one-hour interview is common, growing with the number of speakers and how clean the audio is. That is why the manual route makes sense for short passages and difficult audio, while longer interviews usually go through automatic transcription first.

Can I transcribe an interview on my phone?

You can record on a phone, and some mobile apps offer speech-to-text, but serious interview transcription is easier on a computer, where you can control playback speed, use keyboard shortcuts, and keep a document open beside the audio.

How do I label who is speaking?

Use consistent short labels from the first line — initials or S1/S2 — and fix them to full names in the cleanup pass. Some services attempt automatic speaker separation, but with overlapping speech the labels still need a human review.

Do I need permission to record an interview?

Possibly, depending on where you are and what the interview is for. Consent and recording rules differ between journalism, research, and workplace contexts and between jurisdictions. Check the applicable rules in your location and get any release forms signed before the interview.

Can ToTextly transcribe my interview recording?

Yes. Audio to Text accepts local MP3, M4A, WAV and MP4 recordings up to 4 MB and processes the first 10 minutes. Sign-in is required. Speaker labels need review; download TXT or timed SRT when available. Public YouTube interviews with accessible captions use the separate link tool.

What format should I keep the transcript in?

Plain text as the master copy, with a naming scheme that includes the date and interviewee, stored next to the audio. Export to other formats when a specific tool needs them, but keep the searchable TXT as the source of truth.

Related Guides

How to Transcribe a YouTube Video