Which languages do you support?

28 languages today — each runs the engine mix that hears it best.

How fast is a transcript ready?

An hour of audio is usually ready within 30 minutes. The same recording is transcribed several times over and the readings are cross-checked — slower than a single pass, and that is what makes the uncertain spots findable. Review starts the moment it lands: you check only the marked spots.

How does billing work?

Two ways to transcribe. Transcribe Local runs on your own computer, unlimited for everyone; the merge step is free with your own API key, or $1.20 per hour with ours. Transcribe Cloud is $6.00 per hour pay-as-you-go in all 28 languages, prorated to the exact second; on a plan it drops to $4.80 / $4.20 / $3.60 per hour (Starter / Pro / Max), and the monthly fee is that month’s credit, with the same rate after it runs out. Rate locks at upload; failed jobs are never billed; a job still in the queue can be cancelled for a full refund of the amount held.

Word (.docx) and plain text, with speaker labels.

Can I do more with a finished transcript?

Yes — a confirmed transcript can become a narrative that goes straight into analysis, or a redacted copy that honours what you promised participants. Run one step or both; on Transcribe Cloud transcripts there's no extra charge. One click from the detail page at zero doubts.

How do invoices work?

Cards, PayPal, Apple Pay and Google Pay are accepted; in mainland China, top-ups can also be paid with WeChat Pay (not subscriptions for now). A proper invoice (Transcribe Labs LLC) is issued with every payment — 28 languages, one vendor, one invoice.

What happens to my data?

Audio is auto-deleted 7 days after transcription; transcripts are kept for 30 days and can be deleted any time.

Is my audio uploaded?

With Transcribe Cloud, yes — it is uploaded, transcribed and deleted 7 days later. With Transcribe Local it never leaves your computer: open-source engines transcribe it there, and only the text is sent up to be merged.

Can you transcribe group discussions?

Yes. Tick “This is a group discussion” when you upload and give each speaker a short sample of their voice; every line is labeled with their name, and lines we can't attribute with confidence are marked for you to check. Every language on Transcribe Cloud except Cantonese; on Transcribe Local, every language except French and Cantonese.

Can someone else review it for me?

Yes. Click “Ask someone to review” on the transcript page and send the link to the interviewer or a colleague: no sign-up, they open it, listen and confirm, every spot is recorded under their name, and you see their progress.

What if a transcript isn't accurate?

Email hello@transcribe.solutions within 7 days of delivery, naming the file and the specific places you think are wrong. We check those places against the original audio and reply with our conclusion. Where the defect is serious and ours, that transcription is refunded in full — full procedure in §4 of the Terms.

Stop re-listening to the whole tape.

Several engines hear the same audio and flag every disagreement — you check the marks, and keep the rest of your hours.

Transcribe Cloud

Most accurate

Several commercial engines at once

Names checked online — we handle it all

A 1-hour interview ≈ $6.00

Transcribe Local

Most private

Audio never leaves your computer

Offline edition: the text stays there too

Open-source engines, unlimited

28 languages · every doubt flagged

Start free →

30 free minutes for every new account

Demo · 51:22In review · 1

04:12

Interviewer: How did you handle the coding for that round?

04:19

Interviewee: We coded it in NVivo — everyone on the team already knew it, so it was one less thing to set up.

Engines disagreeChecked online

“NVivo”

Engines heard: in vivo ×2 · Envivo ×1 · NVivo ×1

Web lookup: the analysis software by that name

The count says “in vivo”. Verified: it’s NVivo, the analysis software. Your call.

Confirm ✓Replace

4 engines · 1 web check · 1 to confirm

The problemThe transcript won't tell you where it's wrong

Three old problems no one solves for you.

It won't tell you where it's wrong

A single-pass transcript gives you no way to know which line to trust. The only check is re-listening to everything.

Your hours aren't on the bill

The tool fee is the small part. The real cost is the hours you spend checking — no invoice shows them, so no tool ever tried to save them.

"Looks fine" isn't evidence

A transcript that reads smoothly can still be missing things. If it can't show what was checked, and against what, confidence is just luck.

So the first thing we did was not to transcribe better — it was to show where it's uncertain.

Review firstWe flag every uncertain spot and hand you the call

The transcript that shows its doubts.

We don't sell "trust us." Several engines hear the same audio; every place they argue is flagged with a timestamp — one click replays that second. Only when every doubt is cleared does the transcript say "safe to export."

01Upload02Transcribe03Review only the marks04Export

Review only the marked spots — never re-listen to the whole tape.

Every mark carries a timestamp.

Click one and that exact second plays back. You never hunt for where the line was — the mark is the location.

We can show it because the same audio was heard more than once.

Why it's accurateSeveral engines listen at once — where they disagree is the mark

Several "stenographers" listen at once — then reconcile their notes.

For each language we pick the engines that hear it best, run them in parallel, cross-check every line, and verify key names and places against the open web. Where they disagree is exactly what gets flagged for you. Transcribe Cloud runs commercial engines; Transcribe Local runs open-source ones on your own computer.

At least four engines, in parallel

Every language runs one primary engine plus three or more reference engines, picked for that language. The same sentence is transcribed at least four times.

Four readings, reconciled line by line

Not a majority vote. Wherever the spellings differ, that spot is listed on its own, with what each engine actually heard.

One more pass before delivery

Key names and places checked against the open web, your glossary applied word by word — and whatever is still uncertain is flagged for you.

Hearing it right has two parts: what was said, and who said it.

Who said itEach person leaves a short voice sample first

However many people, however long the recording — nobody gets mixed up.

Many transcription tools start mixing people up once there are several voices and the recording runs long: “Speaker 3” in the first half becomes “Speaker 1” in the second. We have each speaker register a short sample of their voice and recognise them by voice for the whole session. Where we can't tell who it was, we mark it instead of guessing.

Register the voices first

About 10 seconds each, up to 10 people. The round of introductions at the start works best.

Recognised by voice, start to finish

On the same 37-minute, 7-person meeting, leaving it to the transcription engine got fewer than 70% of lines to the right person. Where we give a name, more than 99% are right.

Unsure? We mark it

Crosstalk, or a voice too brief to recognise, gets a dotted line for you to confirm — we don't fill in a name just to have one.

Every language on Transcribe Cloud except Cantonese; on Transcribe Local, every language except French and Cantonese.

Group discussion · 5 voices registered