Live dictation
Text appears while you speak, then is corrected when you stop. If the connection drops, everything transcribed so far is kept and dictation resumes.
Powered by Voxtral, from Mistral AI
Speak or upload a recording: TechVoice transcribes as you talk, fixes the terminology of your field and remembers every correction. Read it over, and it’s filed.
100 free points, about 100 minutes of dictation · no credit card required
Correction of phaluxhallux valgus with Scarf and Akin osteotomies. Post-operative shoe for six weeks.
Your vocabulary
Three simple steps, and a memory that grows with every dictation.
Speak into your computer’s or phone’s microphone, or drop in a recording made elsewhere, up to 60 minutes long.
Text appears as you speak. When you stop, it is corrected within seconds using your vocabulary: technical terms, proper names, acronyms.
Fix a misheard word: the correction is remembered, and TechVoice applies it to your next dictations.
Everything TechVoice learns is set out in plain view in the Memory tab: you stay in control of what it remembers.
Upload sample reports or letters: TechVoice keeps the terms, not the text.
Enter your field: TechVoice looks up its terms in the scientific literature, in the language you dictate in.
Every word you fix becomes a “misheard → correct” pair, applied to your next dictations.
Voice dictation software built for the texts that matter, from the first word to the filed note.
Text appears while you speak, then is corrected when you stop. If the connection drops, everything transcribed so far is kept and dictation resumes.
Voice memo, meeting, recorded consultation: drop in the file, up to 60 minutes and 100 MB. A 30-minute meeting is transcribed in about a minute.
Find out more — Upload recordingsEnter your field and upload a few sample documents (PDF, Word): TechVoice picks out your terminology. The text of the documents is not kept.
Every word you correct becomes a rule for your next dictations. Everything TechVoice learns stays in plain view, and you can edit or delete it.
Turn a dictation into an email, a short message or polished prose: verbal tics disappear, nothing is invented, and anything missing is left in square brackets.
Your dictations become notes, organised in folders and found in an instant with search.
Summary, structured report, to-do list, email or your own instruction: a draft to review, in one click, without touching the note. Handled by Claude, from Anthropic, and only when you run them.
The dictation built into your phone or word processor handles everyday language well. Your profession demands more.
| Keyboard | General-purpose dictation | TechVoice | |
|---|---|---|---|
| Speed | About 35 to 40 words per minute | As fast as you speak | As fast as you speak, with the text shown live |
| Your specialist terms | Accurate, but you have to type them | Often misrecognised | Corrected with your vocabulary |
| Learns from your corrections | Not applicable | Rarely | Yes, with every edit |
| Fixing misheard words | Not applicable | Rarely tailored to your profession | Strictly limited and visible: nothing is reworded |
| Where your text goes | It stays on your device | Depends on the provider | To Mistral AI and Scaleway, in France |
TechVoice adapts to the vocabulary of your field, whatever it may be.
Medical dictation for consultation notes, operative reports and letters: conditions, techniques and devices spelt correctly, without patient-identifiable data.
Find out more — Doctors and surgeonsClient meeting notes, case summaries and first drafts of letters, with the exact legal terminology.
Find out more — Lawyers and legal professionalsClient meeting reports and analysis notes: products, acronyms and regulatory references.
Viewings, inventory reports, mandate notes: dictated on site, filed before you are back at the office.
Engineering, consulting, research… Enter your field and TechVoice learns its terminology.
Your name goes on every report. TechVoice corrects recognition errors, never what you actually said.
Correction only touches misheard words. If it strays too far from what you said, TechVoice keeps the recognised text as it was.
One click shows what the correction changed in the recognised text, so you know exactly what you are reviewing.
Vocabulary, documents and learned corrections are all in plain view, and you can edit or delete them at any time.
Network drop or service outage: text already transcribed is kept. If the correction fails, you keep the raw transcript.
Your voice and your texts are transcribed and corrected by Mistral AI, and hosted by Scaleway, in France.
Full details, provider by provider, are on our Privacy page (in French).
Your audio is used for transcription; after that, TechVoice does not keep it.
From the documents you upload, TechVoice keeps only the terms it extracts.
No analytics, no advertising cookies: only the ones you need to sign in.
Deleting your account erases your notes, your vocabulary and your corrections; encrypted backups expire within 14 days.
One exception: quick actions, when you run them, send the note’s text to Anthropic (Claude) in the United States, under standard contractual clauses. Nothing is sent without your click.
TechVoice does not hold France’s certification for hosting health data (HDS). Do not dictate anything that could identify a patient: no name, no date of birth, no record number.
Start for free, no credit card required. Subscribe once TechVoice has become part of your working day.
To try it out
€0
For everyday dictation
€14 / month
For high volumes
€40 / month
Every plan includes
Prices incl. VAT. About 1 point per minute of dictation, transcription and correction included; other features use points depending on how you use them. At each renewal, your balance is reset to your plan’s allowance: unused points do not roll over. No commitment: cancel at any time. Plans are intended for professionals: see our terms of sale (in French).
The essentials in one minute, without the jargon.
In French and in English: dictation, correction and interface. The flag at the top of the app switches language at any time. Our accuracy measurements cover French. Spanish, German and Dutch are in preparation.
You start for free: signing up gives you 100 points, enough for about 100 minutes of dictation, with no credit card required. After that, there are two subscriptions with no commitment: Essentiel at €14 and Ultra at €40 a month.
General-purpose dictation handles everyday language well, but stumbles over the terms of your profession and rarely learns from your corrections. TechVoice corrects every dictation using your vocabulary and remembers every fix, without rewording what you say.
No. TechVoice runs in your browser, Chrome or Safari, on a computer or a phone. All you need is a microphone.
Yes: transcription and correction run on Mistral AI’s servers. If the connection drops during a dictation, the text already transcribed is kept.
Transcription uses Voxtral, Mistral AI’s family of speech recognition models, and correction uses a Mistral AI language model. We explain this choice and share our measurements in our hands-on review of Voxtral.
From three sources: the field you enter, the sample documents you upload (only the terms are kept) and, above all, your corrections. Every word you fix becomes a “misheard → correct” pair, applied to your next dictations. You can view and edit all of it in the Memory tab.
It is designed not to. It only fixes recognition errors, and an automatic check rejects it if it strays too far from the dictated text, taking word order into account: a misplaced “not” changes everything. You can see what was corrected, and you always review a text before using it.
Not at the moment. TechVoice does not hold France’s certification for hosting health data (HDS): do not dictate names, dates of birth, record numbers or any other information that could identify a patient. This is a commitment set out in our terms of sale (in French).
Your voice and your texts are transcribed and corrected by Mistral AI, and hosted by Scaleway, in France. Audio is not kept after transcription. Sign-in is handled by Clerk and payment by Stripe, with possible transfers of your email address and name outside the European Union. Everything is detailed on our Privacy page (in French).
The note’s text and your instruction are sent to Anthropic (Claude), in the United States, under standard contractual clauses, and only when you click. Anthropic does not train its models on this content. The note itself is left unchanged: you get a draft to review.
Text already transcribed is kept and TechVoice reconnects to carry on. If the correction fails, you keep the raw transcript, clearly marked as such.
Yes: an audio file of up to 60 minutes and 100 MB (m4a, mp3, wav, ogg, flac…), such as a voice memo from your phone or a meeting recording. It is transcribed and corrected just like a dictation.
About 1 point per minute, transcription and correction included. The 100 free points you get when you sign up cover about 100 minutes; paid plans renew your balance every month. See pricing.
Yes, from your account and at no cost: your subscription ends at the close of the period you have paid for, and your remaining points can be used until then.
Speaking is three times faster than typing, but the real gain is won or lost at review. A worked calculation and cited studies to save time on your reports.
Models, languages, pricing, open weights: all about Voxtral, Mistral AI’s speech recognition, and what we measured when using it for professional dictation.
Speech recognition, AI correction, correction memory: how modern voice dictation works, and why it stumbles over the technical terms of your profession.
100 free points, no credit card required. Your vocabulary starts taking shape from your very first dictation.
Try it free