How to transcribe an interview

2026-10-13 How to transcribe an interview

Transcribing an interview by hand is one of those jobs everyone underestimates. The rule of thumb people quote is four to six hours of typing for one hour of audio, and that's if the recording is good. If it isn't, you'll be rewinding the same ten seconds of mumbling twelve times.

The good news is you don't have to type it anymore. The better news is that the part that actually matters, the cleanup, is where you can save the most time by being a bit smart about it. Here's the workflow that works for me, and it applies whether you're a journalist, a researcher, a student with a class project, or someone who just recorded a chat with their grandfather.

Start before you press record

Most transcript problems are recording problems in disguise. A few things worth doing up front:

Test the audio for thirty seconds. Play it back with headphones. If you can't hear the person clearly now, nobody will later, including the software.
Put the mic close to the person talking. Closer matters more than fancier. A phone on the table between two people is okay; a phone held near the guest is better.
Ask people to say their name and the date at the start. It sounds silly, but it saves you from wondering who is who in a file named Recording 47.m4a.
Get consent on tape. If you're recording someone, ask on the recording itself. It protects you and it's just polite.

If it's a video call, record the audio locally if you can rather than relying on the call software's compressed version. The difference in clarity is real.

Decide what kind of transcript you need

This is the step people skip, and it changes how much work you have ahead of you.

Verbatim means every "um", "you know", false start and laugh gets written down. Researchers doing discourse analysis and lawyers sometimes need this. Most of us don't.

Clean (or "intelligent") verbatim drops the filler and the stutters but keeps the meaning and the speaker's wording. This is what you want for articles, quotes, case studies and general notes. It reads like a person talking, minus the noise.

Summary notes are just the key points. Not really a transcript, but sometimes all you need.

For most interviews, go with clean verbatim. If you're not sure, start verbatim, because deleting words from a full transcript is easier than trying to put them back from memory.

Run the audio through a speech-to-text tool

This is the part that used to eat the weekend. Open an audio to text tool, drop in the file, and let it work. An MP3, M4A, WAV or a video recording all work, since the tool only cares about the audio track. If your file is a phone recording, voice memo to text is set up for exactly that.

A couple of practical notes:

Use the original file. A recording that has been re-compressed several times (sent through a messaging app, then converted, then sent again) loses detail, and the transcript pays for it.
For a long interview, use a laptop or desktop and keep the tab open. Phones can do it, but they're slower and tend to throttle when the screen locks.
If the transcript will be used for something that matters, pick the more accurate model when you have the choice. Slower is fine when the output is going into a published piece.

Expect a first draft, not a finished document. Modern speech recognition is good on clear English speech, and it's honestly impressive, but it will still trip on names, numbers, jargon and anyone speaking over someone else.

Add speaker labels yourself

Here's something worth being upfront about: most free tools, including this one, give you the text but don't reliably work out who said what. For a two-person interview that's actually an easy fix.

The simplest approach is a find-and-label pass. Read through while the audio plays, and at every change of speaker, start a new paragraph with the name:

Maya: So how did you get started?

Tom: Honestly by accident. I was fixing my brother's website.

A few conventions that make life easier:

Use first names or initials, and be consistent.
Start a new paragraph every time the speaker changes, even for a one-word answer.
For interviews with one interviewer, "Q:" and "A:" is perfectly fine and quicker to type.
If two people overlap, note it in brackets, like [both talking], and move on. Nobody wants to untangle it word by word.

If you have more than two speakers, such as a panel or a focus group, set aside extra time. This is the stage where human attention still beats software.

Do the cleanup pass with the audio playing

This is the step that makes the transcript trustworthy. Don't just read the text on its own. Play the audio at 1.25x or so, and follow along with the text.

What to look for, roughly in order of how often it goes wrong:

Names. People's names, company names, place names. Check the spelling against LinkedIn, a website, or the guest themselves.
Numbers and dates. "Fifteen" and "fifty" sound alike. So do "2018" and "twenty eighty". Anything you might quote needs a double check.
Technical terms and jargon. The software has never heard of your niche's product names, so expect creative spellings.
Missing negatives. "I do recommend it" versus "I don't recommend it" is a tiny difference to a speech model and a huge one to a reader.
Sentence breaks. Add punctuation where the meaning needs it. Break up giant paragraphs so someone can scan it.

You'll quickly find that you don't need to listen to everything at full attention. Skim the easy parts, and slow down on any sentence that feels slightly off when you read it. Those tend to be the mistakes.

If you want to read more about getting better text from the start, our tips for accurate transcripts cover the recording side in more detail.

Add timestamps where they actually help

Timestamps are useful for two things: finding a moment again, and syncing text to video. For a normal interview transcript you don't need one on every line. A timestamp every few minutes, or at each new topic, is plenty. It lets you jump back to "the part about the budget" without scrubbing through the whole file.

If you need subtitles instead, that's a different format with its own rules, and we've written a separate guide on turning audio into SRT subtitles.

Format it so you can actually use it

A transcript that's one giant wall of text is technically correct and practically useless. Spend ten minutes on layout:

Put a header at the top: interviewee, interviewer, date, location or platform, and the file name of the recording.
Keep speaker labels bold or in a consistent format so your eye can find them.
Highlight (or bold) the quotes you'll likely use, so you're not rereading the whole thing later.
Save the original audio next to the transcript with a matching file name.

If you're a student or researcher, check your department's convention. Some want line numbers, some want a specific header, and a few want everything verbatim with pauses marked. Ask early, because reformatting at the end is miserable.

Privacy, because it's a real consideration

Interviews often contain things people said in confidence: names, health details, opinions about their employer. Before you upload a recording to any service, ask where the file goes and who can see it. Some services keep recordings for training or review. Others delete them after processing.

The free mode of this tool runs the speech model on your own device, so your file isn't uploaded anywhere. That's not the point of this guide, but if you're handling sensitive interviews, it's worth knowing where your audio goes whichever tool you choose. For anything covered by a research ethics agreement or a source-protection promise, read the terms before you press the button.

A realistic timeline

For a one-hour interview with decent audio: a few minutes for the first draft, 30 to 60 minutes for the cleanup pass with the audio playing, and about 10 minutes for labels and formatting. That's roughly an hour and a half instead of most of a day.

The short version

Record close and clean, decide on clean verbatim unless you have a reason not to, let software do the first draft, then spend your time where humans are still better: names, numbers, negatives and who said what. That's the whole job.

When you're ready, drop your file into the audio to text tool and see how the first draft looks. It's usually closer than you'd expect.