Interview Transcription

What an interview transcript looks like

Below is the unedited output for a public interview recording: speaker labels and timecodes exactly as SozAI returned them, before anyone corrected a word.

Try it on your own recording first

The box at the top of this page transcribes the first five minutes of an interview file, free and without an account. Your browser cuts a longer recording on your device, so only those five minutes are uploaded. The transcript comes back split into speaker turns, and once you have seen it, it is deleted at the speech recognition service; we keep neither the audio nor the text.

It takes MP3, M4A, WAV, OGG/OPUS, AAC, FLAC, WEBM and MP4, in files up to about 30 MB on a computer and 12 MB on a phone. Five minutes is enough to see whether the speaker labels hold on your audio. A call recorded on speakerphone works, and so does a conversation across a café table, with more mistakes on the quieter voice.

Name the speakers once

Every file and every recording made in the app comes back with speaker labels: Speaker A, Speaker B and so on. Rename them in the iPhone or Android app or on web.sozai.app and the new names replace the letters in that transcript on both, and in your exports.

Two habits save time later. Use roles rather than real names when the person should not be identifiable ("Interviewer", "Participant 3", "Source"), because the names you type are stored with the transcript. And check the first two minutes: when two voices sound alike, or one person only says "yes" and "right", the labels can swap at short interjections.

Speaker renaming is not in the Mac and iPad app yet. Rename on the phone or the web.

Checking a quote against the audio

A transcript is a draft of what was said. The recording is the record. Before a sentence goes into print or into your findings, play it.

In SozAI every passage carries the time it starts. Tap or click the timecode and playback starts at that line, in the phone app and on the web. Listen to the sentence before and after it as well: a quote that is word-perfect can still change meaning when you cut the question that came before it.

If the transcript got a word wrong, fix it in the iPhone or Android app, line by line or the whole text at once. The web and Mac versions show the text but do not edit it yet. Your correction is saved to the transcript itself, so the corrected version is what you export and share.

Names and numbers are where automatic transcripts slip most often. Look for each of them in the transcript and play it back.

Interviews in other languages, or in two

SozAI detects the language from the audio, so you never pick it. That covers 99 languages, and it means a German interview and a Turkish one can sit in the same folder without any setting changed.

Quality is not the same in every language. We measured it on 200 clips in 10 languages and published the method and the result for every clip in our accuracy test. Those were short read sentences in quiet rooms. A real interview, with crosstalk and a café behind it, will do worse.

Detection happens once per recording. An interview that runs in one language and drops into another for a few sentences is transcribed in the main language, and those sentences can come out wrong; check them against the audio. If the conversation switches for good halfway through (an interpreter takes over, say), splitting the file at that point gives each half its own detection.

Need colleagues to read it in their language? Translate the transcript in the app or on the web and keep the original next to it. Quote from the original.

For journalists

On deadline you need two things from a transcript: a quote you can prove, and a source you have not exposed.

The proof part is mostly speed. The transcript is text, so you look for the word you remember (Ctrl+F works on the web) instead of scrubbing through an hour of audio. The timecode takes you to the second it was said, and the audio stays in your account for as long as your audio setting keeps it. Field audio works if the phone is close to the person talking. A phone on the table between you is fine; a phone in your pocket in a loud street is not.

Keeping the source safe is about where the recording goes. When you upload it, the audio and the transcript are stored on our servers, and the speech recognition runs at a partner company in the US. Every company that touches the file is named on our sub-processor list. You can delete a recording, or your whole account, yourself; that removes the audio, transcript and notes from our servers.

Our advice: when the identity of a source is itself the secret, do not upload that recording anywhere, to us or to anyone else. Transcribe it on a device you control.

For researchers

Research interviews come with conditions a newsroom does not have: consent forms that say where audio may go, a transcription convention you have to keep, and software you will import into. We wrote those up separately in qualitative research transcription, including how a SozAI transcript gets into NVivo, ATLAS.ti or MAXQDA and what to tell an ethics board.

Hiring and customer interviews

If you interview customers for product research, rename the labels to "Customer" and "Interviewer" and the transcript reads like a script. Then ask the recording a question in the app ("what did she say about pricing?") instead of scrolling for it. The same works across a hiring loop: ask every candidate the same questions and the transcripts can be compared side by side.

The first 30 free minutes cover about one interview. That is enough to see whether the speaker labels hold up on your own audio before you pay anything.

Tell the person you are recording, at the start and on the recording. For job candidates, check your company's policy on keeping interview recordings and how long you may hold them.

Export formats, by platform

SRT files never carry speaker names; use VTT when you need both names and timing. There is no Word export: open the TXT in Word, or send the PDF.

SozAI, Otter, TurboScribe, Notta and Rev for interviews

Prices and limits as each company publishes them, checked 25 October 2026. Click a name for its pricing page; cells marked otherwise link to the page they come from.

Our pick, honestly: if your interviews are English video calls, Otter is built for that and joins the call for you. If you have a backlog of long files and want one flat price, TurboScribe's paid plan is hard to beat. If a quote will be printed and you cannot afford a single wrong word, pay Rev for a human-verified transcript. SozAI is the better fit when you record in the room or on the phone, when the interviews are not all in English, and when you want to move between phone, laptop and browser on one account.

Who SozAI suits for interviews, and who it does not

A good fit if you:

• record interviews in person or on the phone and want the transcript on your laptop by the time you sit down
• interview in more than one language, or in a language the big English-first tools leave out
• need timecodes to go back to the audio and check every quote
• want a summary of the interview, and answers to questions about it, without pasting it into another tool

Look elsewhere if you:

• need a certified or human-verified transcript (a human service such as Rev does that)
• want a bot that joins Zoom or Teams and records the call for you
• need a Word file with the speaker names already formatted
• work under a contract that requires an independent security audit report before any file is uploaded

Interviewing as a team

Today each person on a team has their own SozAI account. There is no shared workspace or admin role yet, and no sign-in through a company identity provider. Sign-in is with Google, Apple or a one-time email code. Team and Education seats are bought by card at web.sozai.app, $12.99 a month or $99.99 a year per seat, with an invoice in your organization's name: how buying for a team works. More on how teams use it on SozAI for teams.

Where your interviews go

• Recordings travel to us over HTTPS and are stored on our servers.
• Speech recognition is done by a partner company in the US. Summaries, notes and answers to your questions are written by AI models from other providers. Every company that processes your data, and what it receives, is named on our sub-processor page.
• We do not sell your recordings, and our own code never uses them to train models.
• Deleting a recording, or your whole account, removes the audio, the transcript, notes, cards and quizzes from our servers. Encrypted database backups are kept for up to 12 months. Deleting in SozAI does not delete the copy held by the speech recognition partner.
• What we do not have yet: an independent security audit report, sign-in through your company's identity provider, and team admin roles. Details on security at SozAI.

Price

The first 30 minutes are free, once, with no card. After that, Premium is $4.99 a week or $79.99 a year, billed by the App Store or Google Play, with up to 700 hours of transcription. See pricing.