Transcripts you can check.
Upload audio or video and get back a transcript with punctuation and a timestamp on every word — in 82 languages. Any line plays back the second it came from. Fix what the machine got wrong.
Free to try. No account, no card.See pricing
Weekly sync
Demo · no audio0:04 / 0:38 · 2 speakersSpeaker 1
Numbered speakers, not names — rename one and it carries through. An illustration, not a recording.
Upload recordings straight from the apps you already use
- ZoomMP4 · M4A
- Microsoft TeamsMP4
- Google MeetMP4
- Voice MemosM4A
- LoomMP4
- 中文Chinese
- English
- 한국어Korean
- FrançaisFrench
- ItalianoItalian
- РусскийRussian
- Tiếng ViệtVietnamese
- Bahasa IndonesiaIndonesian
- TürkçeTurkish
- NederlandsDutch
- УкраїнськаUkrainian
- ShqipAlbanian
- ՀայերենArmenian
- AsturianuAsturian
- EuskaraBasque
- БългарскиBulgarian
- CatalàCatalan
- ČeštinaCzech
- EestiEstonian
- SuomiFinnish
- ქართულიGeorgian
- ગુજરાતીGujarati
- עבריתHebrew
- ÍslenskaIcelandic
- ಕನ್ನಡKannada
- ខ្មែរKhmer
- ລາວLao
- LietuviųLithuanian
- МакедонскиMacedonian
- MaltiMaltese
- मराठीMarathi
- नेपालीNepali
- NorskNorwegian
- فارسیPersian
- RomânăRomanian
- SlovenčinaSlovak
- KiswahiliSwahili
- தமிழ்Tamil
- OʻzbekchaUzbek
- Wolof
- YorùbáYoruba
- 粵語Cantonese
- 日本語Japanese
- EspañolSpanish
- DeutschGerman
- PortuguêsPortuguese
- हिन्दीHindi
- ไทยThai
- Bahasa MelayuMalay
- العربيةArabic
- PolskiPolish
- Afrikaans
- አማርኛAmharic
- অসমীয়াAssamese
- AzərbaycancaAzerbaijani
- বাংলাBengali
- မြန်မာBurmese
- HrvatskiCroatian
- DanskDanish
- Filipino
- GalegoGalician
- ΕλληνικάGreek
- Hausa
- MagyarHungarian
- Basa JawaJavanese
- Қазақ тіліKazakh
- КыргызчаKyrgyz
- LatviešuLatvian
- LëtzebuergeschLuxembourgish
- മലയാളംMalayalam
- Te Reo MāoriMaori
- МонголMongolian
- Sesotho sa LeboaNorthern Sotho
- ଓଡ଼ିଆOdia
- ਪੰਜਾਬੀPunjabi
- СрпскиSerbian
- SlovenščinaSlovenian
- SvenskaSwedish
- తెలుగుTelugu
- CymraegWelsh
- isiXhosaXhosa
- isiZuluZulu
Take it out as
- languages
- 82languagesThe picker lists 110 variants
- variants with speaker labels
- 110variants with speaker labelsUp to 2 h in 96 of them
- timestamps
- Word-leveltimestampsClick a line to hear it
- longest file on Pro
- 8 hlongest file on ProOne upload, one transcript
Upload it, read it, check it.
Fix it yourself, fast.
No recogniser gets names, jargon or crosstalk right every time. Click a paragraph and correct it — your edit doesn't break the timings. Rename a speaker once and the name follows the whole transcript.
Every word timed
Checking a quote takes a second, not a scrub.
Click any line to hear it. The transcript follows as it plays, so checking a quote takes a second rather than a scrub.
Speaker 2
We never voted on the second tranche. It was added to the budget after the committee had already signed off.
Speaker 1
Who added it?
Speaker 2
I'd rather not say that on the record.
- The word being spoken is marked as it plays
- Click a word to play from exactly there
- Timecodes sit in their own column, so they never jitter
Yours to correct
Fix a misheard name once.
Click a paragraph and fix it. The timings survive your edit, and a custom word list stops the same mistake twice.
- Offers to make the same fix everywhere else — and says how many places first
- Split a paragraph, merge two, or move a line to the right speaker
- Every edit can be undone, and it saves as you go
Try it without an account
Upload or record something and read its transcript, free and with no account.
Every claim keeps its source.
Summaries are drawn only from what was said. Answers come back with the moment they came from. So when it matters, you check it in two seconds instead of taking it on faith.
Speaker 2
We never voted on the second tranche. It was added to the budget after the committee had already signed off.
Speaker 1
Who added it?
Speaker 2
I'd rather not say that on the record.
Speaker 1
Could the chair have added it on her own?
Speaker 2
Not without the finance office knowing about it.
Did she say who added the second tranche?
No. She says it was added after the committee had signed off, and declines to say who added it.
Sources12:0412:14
CopyRegenerate
An illustration, not a real recording or answer.
- Cites the moment
- Each answer lists the moments it drew on, as timecodes. Click one and the recording plays from there.
- Only from the transcript
- Answers come only from what was said, not from what a model happens to know about the subject.
- Says what it didn't read
- If a recording is too long to read in one go, the answer says how much of the middle it didn't see — instead of letting you assume it read everything.
It knows what you recorded.
A sales call and a user interview need different notes. Pick what you recorded and the summary is shaped for it.
Get the quote right.
Themes, quotes worth keeping, and claims worth checking.
It's told not to smooth things over.
Each kind of recording gets its own instructions. For a meeting, the summary is told to leave open questions open; for an interview, to quote people word for word; for user research, to record a request as a request, not as a recommendation. It is still a model, so check what matters against the recording.
Budget committee: the second tranche
A committee member describes how a second tranche entered the budget after committee sign-off, and declines to say who added it.
Themes
- How the second tranche entered the budget
- Who had the authority to add it
Notable quotes
- “We never voted on the second tranche.”
- “I'd rather not say that on the record.”
Claims to check
- The tranche was added after the committee had signed off
- The finance office would have known about it
An illustration of the summary format, not a real recording.
Translate it, export it, share it.
Every plan is the whole product, not a demonstration of it: summaries, translation, questions, share links and every export format are on all of them.
Translation in place
The first version of the book was six hundred pages long.
这本书的第一版有六百页。
La première version du livre faisait six cents pages.
その本の最初の版は600ページもありました。
Die erste Fassung des Buches hatte sechshundert Seiten.
La primera versión del libro tenía seiscientas páginas.
And how many were left?
那最后剩下多少页？
Et combien en est-il resté ?
それで、何ページ残ったんですか？
Und wie viele sind übrig geblieben?
¿Y cuántas quedaron?
Two hundred. And I still think fifty of those are too many.
两百页。而且我还是觉得有五十页是多余的。
Deux cents. Et je trouve toujours qu'il y en a cinquante de trop.
200ページです。それでも、まだ50ページは多すぎると思っています。
Zweihundert. Und ich finde immer noch, dass fünfzig davon zu viel sind.
Doscientas. Y todavía creo que sobran cincuenta.
Export
Plain text, subtitles or a Word document — straight from the transcript.
Share a read-only page
Anyone with the link can read it, without getting access to your account.
Where it falls over.
Every transcription tool has these. Most don't mention them.
Recordings that switch language
Auto-detect has to decide what language each stretch of speech is in. Where a recording switches language, what was said in the other one may come back translated instead of transcribed, or not come back at all — and nothing is marked as missing, so the transcript will look complete. Name the main language if you know it.
Speaker labels stop at 2 hours in 96 variants
In 96 of the 110 variants, the speakers are labelled by a separate pass, which takes recordings of up to 2 hours. A longer recording in one of them comes back with full timings and punctuation, but without speaker labels — and so can a shorter one if that pass fails. Either way, the transcript says so.
Long recordings drift
On long recordings, speaker labels can wander from one section to the next — a consequence of how long audio is processed, not a setting. We wrote up why.
Pricing
Everything the product does is in every plan. The only thing you are buying is time.
Free
Free
Keep your transcripts, with every feature on them.
- Keep a transcript bought on the trial page in your account
- Speakers, word-level timestamps and editing
- Summaries, translation, questions and sharing
Pro
US$24.99/mo
For people who transcribe every week.
- 30 hours a month
- Files up to 8 hours
- Everything in Free, without the ceiling
Before you upload.
Anything not here, write to support@txscribe.ai.
Can I try it without an account?
Yes. Put one recording of up to 10 minutes through it — upload it or record it — and you get a real transcript back, with no account and no card, and can read all of it. It is kept for 48 hours; get Pro and you can keep all of it in an account.
What if the transcript is wrong?
Then you fix it — that's the part we built most carefully. No recogniser is perfect on names, jargon or people talking over each other. The editor keeps your timings, and a custom word list stops the same mistake recurring.
Which languages, and which get speaker labels?
82 languages. The picker lists 110, because some of them come in more than one regional variant. Speaker labels are offered in all of them. In 14 of them (Chinese (Mandarin), English (US), English (UK), English (India), Japanese, Korean, Spanish (Spain), French (France), German, Italian, Portuguese (Brazil), Hindi, French (Canada) and Spanish (US)), the recogniser labels the speakers itself; in the other 96, a separate pass does, on recordings of up to 2 hours. A longer recording in those comes back without labels, and so can a shorter one if the pass fails; either way, the transcript says so.
What if a recording switches between languages?
Name the main language rather than leaving it on automatic. Automatic detection has to decide what language each stretch of speech is in, and where a recording switches language, what was said in the other one may come back translated instead of transcribed — or not come back at all. Either way nothing is marked as missing: the transcript will look complete.
What happens to my recordings?
They're stored so you can play them back beside the transcript, and deleted with the transcript when you delete it. They are never used to train anything.
How much does it cost?
Trying one recording costs nothing: up to 10 minutes are transcribed with no account and no card, and you can read all of the transcript. Pro is for people who transcribe every week. See pricing.
Useful before you sign up.
Notes on building a transcription tool, including where ours falls over — and 5 tools that need no account.
- Background noise removerGet a recording back with the hum, hiss, traffic and chatter turned down. Hear before and after, then download a WAV.BeforeAfter
- SRT to VTT converterConvert subtitles between SubRip and WebVTT, or either one to plain text, and shift the timing.SRT00:00:04,000 --> 00:00:07,500VTT00:00:04.000 --> 00:00:07.500
- Word error rate calculatorCompare a transcript against the correct text: the error rate, the alignment, and which errors landed on numbers, names and negations.REFwe did not approve itHYPwe did not approve it↳ a dropped negation
- Transcript cleanerTake the filler words, timestamps and cue numbers out of a transcript, join each speaker's turns and format the names, then copy or download it.INUm, so, we, like, start at nine.OUTWe start at nine.
- Speech time calculatorWords to minutes at a slow, average, fast or typed pace, minutes back to words, and reading time. Paste the text to count it.750 ÷ 150= 5:00
8 October 2026 · 5 min read
How to calculate word error rate, with a worked example
The WER formula is one line. Getting a number you can trust is the rest — aligning the two transcripts, deciding what counts as the same word, and knowing why the result can go past 100%.
8 October 2026 · 6 min read
Why your transcript repeats the same phrase over and over
Sometimes a speech recogniser stops transcribing and writes one short phrase again and again. Why it happens, how to spot the quieter versions of it, and what to do with the part of the recording it swallowed.
26 September 2026 · 6 min read
AI transcription hallucinations: what the research found, and how to catch them
Speech recognisers sometimes write sentences nobody said. What the best-known study measured, what the widely shared numbers do and don't mean, and a practical way to catch invented text before it reaches anyone.
Judge it on your own recording.
Ours is the only transcript we can vouch for, and it isn't yours. Run something real through it — that's the only test that means anything.
Free to try. No account, no card.