Turn French meetings, interviews and lectures into searchable text with speaker labels — Metropolitan, Québécois, Belgian and West African French.
iOS and Android. 30 minutes free, no card.
Does SozAI transcribe French?
SozAI transcribes French audio and video into text across Metropolitan, Québécois, Belgian and West African French. It labels each speaker automatically, timestamps every word, restores accents and cedillas, and exports to TXT, PDF, SRT and VTT. Recordings travel over HTTPS and are stored on our servers in Germany, and every account gets 30 minutes, once per account, without a card.
Français, written the way it is actually written.
Specification
| Native name | Français |
|---|---|
| ISO 639-1 code | fr |
| Writing system | Latin |
| Text direction | Left to right |
| Speakers worldwide | ~310 million |
| Varieties covered | Metropolitan, Québécois, Belgian, West African |
| Speaker labels | Yes — automatic diarization |
| Timestamps | Word level |
| Summaries | Yes, in the same language |
| Export formats | TXT, PDF, SRT, VTT, JSON |
| Maximum file size | 500 MB |
| Free tier | 30 minutes once, no card |
| Where data is stored | Stored on our servers in Germany, encrypted in transit |
| Public transcripts in this language | 749 |
Workflow
Record straight into the app, upload a file up to 500 MB, or paste a YouTube link. SozAI reads m4a, mp3, wav, aac, mp4 and mov.
Choose French when you know the variety, or leave detection on. Recordings that mix French and English — routine in Montréal and in Brussels institutions — stay in a single transcript.
The transcript arrives with speaker labels and word timestamps. Search it, ask it questions, generate a summary in French, or export SRT and VTT subtitles.
Field notes
Liaison and elision glue French together: "les amis" is pronounced as one unit, "je ai" becomes "j'ai". The acoustic signal simply does not contain the spaces the written language requires. A French transcript is therefore always partly a grammatical reconstruction, which is why models that understand syntax outperform ones that only match sounds.
French is unusually rich in words that sound identical and are spelled differently: est / et / es / ai, and the whole -er / -é / -ez family of verb endings. Choosing between them requires knowing the grammatical role of the word, not what it sounded like. This is where most of the remaining error rate in French transcription lives.
Québec French differs from Metropolitan French in vowel quality, vocabulary and rhythm enough that a model tuned on Paris broadcasts degrades on a Montréal conversation. It is treated as a distinct variety rather than an accent, and joual-heavy informal speech remains the hardest case.
Dropping a diacritic changes the word: "ou" and "où", "a" and "à", "sur" and "sûr". SozAI writes the full orthography including accents, the cedilla and the œ ligature, because a transcript that strips them cannot be pasted into a document, a subtitle file or a search index without a manual pass.
From the archive
749 public French transcripts so far — every one produced by the same engine described above. Read them in full, no account needed.
La méthode infaillible pour TOUT apprendre avec l’IA — Transcript
UFC 268 – Une semaine avec Nassourdine Imavov : Mission Top 15…
Construisons une meilleur image des Arabes ensemble — Transcript
Éjaculation précoce : la vérité que personne ne t’explique ! — Transcript
« 17 accusés en taule ! » : le procès historique du…
Petanque Tirer haut, en cloche, souple ou fort — TranscriptSozAI transcribes 99 languages. These have a page of their own.
Answers
Short answers you can quote. Longer ones live in the sections above.
Metropolitan, Québécois, Belgian and West African French. You can pick the variety or let automatic detection choose from the opening seconds. Accuracy is highest on clear, standard speech; heavily informal Québécois and fast overlapping conversation are where you should expect to correct the most.
Yes — é, è, ê, ë, à, ù, û, ç and the œ ligature are all written properly. This is load-bearing rather than cosmetic: several of the most common French words differ only by an accent, and a transcript without them is not directly usable in a document or an SRT file.
Yes. Diarization operates on the audio, so speakers are labelled and every turn timestamped regardless of language. For journalism and qualitative research this is what turns a recording into a quotable source rather than an undifferentiated block of text.
They stay in a single transcript, with each fragment written in its own language and in the order it was spoken. This is the normal situation in Montréal workplaces and in Brussels institutions, and splitting the file by language loses the thread of the discussion.
Every account gets 30 minutes of transcription once, at no cost and without a card. Beyond that, Premium is a subscription through the App Store or Google Play — from $3.99 a week or $44.99 a year in the US — and the price is the same in every language. Current plans are listed on the pricing page.
Where to use it
Everything below works in French, with the same speaker labels and word timestamps.
Transcribe your first French recording
Record, upload a file or paste a link. Accents, speakers and timestamps come back with it.
Free on iOS and Android. No account required for YouTube transcripts.
Point your camera
Scan the code to open the download page on iOS or Android. 30 minutes of transcription free, no card.
