Transcripts

99.99% accurate transcripts.

Ninety-nine percent sounds like a finished job. On a five thousand word transcript it leaves fifty words wrong — names, places and numbers, spelled confidently enough that nobody catches them. We deliver fewer than one, and we guarantee it: if a transcript comes back short of 99.99%, we re-do it free.

What one percent actually costs

Accuracy percentages are easy to skim past. Put them in words instead. Take a transcript of 5,000 words — about thirty-five minutes of conversation — and count what comes back wrong.

99% accurate

The industry standard

50

wrong words in 5,000

Fifty words in every five thousand are wrong. Across a one-hour interview that is roughly ninety mistakes, scattered anywhere.

99.99% accurate

What we deliver

<1

wrong words in 5,000

One error in every ten thousand words. Across the same one-hour interview, expect about one — and we guarantee it.

What a wrong word looks like

Errors are rarely the difficult vocabulary. They are ordinary English words standing where a name should be — which is exactly why they survive a proofread.

Shin Gen Z

Shenzhen

Brian Armstrong Full

Brian Armstrong

the FDD ruling

the FDA ruling

Compared with Rev

Rev is the name most people reach for, so it is the fairest thing to measure against. Same price per minute, four times faster, and a different thing arriving at the end of it.

 SupermixRev, humanRev, AI
Guaranteed accuracy99.99%99%96%
Wrong words per 5,000Fewer than 150200
Price per minute of audio$1.99$1.99$0.25
Standard turnaround180 minutes12 hoursMinutes
Rush turnaround90 minutes, for a rush feeAvailable, for an added fee
Names and terms checked against published sourcesEvery one, every episode
Speaker labelsReal names throughoutNumbered by defaultNumbered by default
Your terminology carried across episodesYes, and it grows each time
Unclear audioMarked, never guessedMarked inaudibleBest guess
Word-level timestampsOn every word

Rev figures are their own published ones as of August 2026, taken from rev.com. An em dash means the point is not covered either way in their published service description, not that it never happens. Wrong-word counts are the accuracy figures applied to 5,000 words.

Every name checked, not guessed

The words a transcript gets wrong are almost never the difficult ones. They are the names — your guest's, your company's, the product you spent the episode talking about. And they come back spelled confidently, in real English, which is why nobody catches them before it publishes.

What gets checked

PeopleCompaniesProductsPlacesAcronymsIndustry jargon
Which words get checked is not a list you have to supply. Anything in the recording that a machine is likely to mishear is pulled out and put through the same process.

What that means for your transcript

Four things you can rely on, on every file you send us.

Checked against the record

Every proper noun in your recording is looked up and confirmed against published sources before the transcript ships. Not sounded out, not inferred from context — corroborated, with the sources cited.

Fixed everywhere, not once

When a spelling is wrong, the correction lands on every occurrence across the whole transcript. A name that appears forty times is right forty times.

Left alone when it is already right

An unusual spelling that happens to be correct stays correct. We do not flatten a real name into the more common one that sounds like it.

Remembered for next time

Confirmed names and terms are kept for your show. By your third episode we already know your co-host's surname, your product names and your industry's acronyms, and get them right the first time.

What you actually get back

The right words

Catches the errors that sound right

The damaging mistakes are the confident ones: a real name mangled into other real words, an acronym a single letter off. They pass every spellcheck. They do not pass here.

Nothing invented

Where the audio genuinely is not clear, the transcript says so. You get a marked inaudible, not a plausible guess you would never think to question.

The right speakers

Real names, not Speaker 1

Everyone in the room is labelled with their actual name, all the way through. Nothing to find and replace afterwards.

Every line with the person who said it

Where a stretch of conversation was attributed to the wrong person, it is checked against the audio and put back where it belongs — so quoting from the transcript is safe.

Ready to use

Verbatim, not edited

Meaning and voice preserved. Nothing paraphrased, nothing summarised, nothing quietly tidied away. What was said is what you get.

Every format you need

Subtitles, documents and data: SRT, VTT, .docx, Markdown, Notion, Descript and JSON. Word-level timestamps throughout, so any line points back to the second it was said.

Multi-mic sessions handled

Recording each speaker to their own track makes for a better transcript, not a harder one. Send the separate files and they come back as one clean timeline.

How it works

There is no workflow to learn. You send a file and you get a transcript back.

The whole process

1. Send us the file

Audio or video, any common format, any length. One file or a folder of them. Nothing to configure, no word list to prepare, no settings to weigh up.

2. We take it apart

Every name checked, every speaker settled, every uncertain word either resolved against evidence or honestly marked. You do not hear from us in the meantime.

3. Back in 180 minutes

Finished and ready to publish, quote or search. Need it sooner — add the rush fee and it comes back in 90 minutes instead.

Pricing

One rate, billed on the length of the audio. No minimum order, no subscription, and no separate charge for any of the work above — the accuracy is the product, not an upgrade.

$1.99

per minute of audio

Standard turnaround

Back within 180 minutes

Included

Rush turnaround

Back within 90 minutes

Rush fee

High volume

Regular shows and back catalogues

Talk to us

Running a weekly show or a back catalogue to work through? Get in touch for bulk pricing.

Common questions

Send us one file and read the difference