French Transcription: Video & Audio to Text

How French speech transcribes

French speech — including Québécois and African French accents — transcribes well. Elisions and liaisons (l'homme, c'est) are written correctly in most cases.

What a French transcript looks like

French transcripts include accents and standard punctuation. Elisions (l', j', qu') are written as in normal French. In casual speech, dropped 'ne' in negations is usually transcribed as spoken, so the text reflects how people actually talk. Numbers are generally written as digits, using the French decimal comma.

Examples

Costs: 75 credits (7.5¢) and 22 credits (2.2¢), including a translated transcript if you select another language. Audio files go in the audio to text workspace.

Silent endings: the main source of errors

French writes many letters it doesn't pronounce, especially at the ends of words. Parle, parles and parlent sound the same; so do été, était, étais and étaient; so do the past participle and the infinitive (mangé, manger). A speech recognizer hears one sound and has to infer the grammar, so French transcripts tend to contain agreement errors that a spell-checker passes, because every word exists.

a / à

"has" versus "to" or "at"

ou / où

"or" versus "where"

ces / ses / c'est / s'est

four spellings of one sound

-é / -er / -ez / -ait

participle, infinitive, imperative or imperfect

leur / leurs

singular or plural possessive

After correcting names, run a grammar checker built for French over the transcript; it flags many of these faster than reading does. Agreement that depends on something said several sentences earlier still needs a person, as does the gender of a past participle after avoir when the object came first (les lettres qu'il a écrites).

Space before the colon: French typography in subtitles

Typography in France puts a narrow non-breaking space before ; : ! and ? and inside « guillemets ». Check what your transcript uses. If you add these spaces when editing captions, use a non-breaking space; an ordinary space lets the player wrap the line and leave a lone ? at the start of the next one. Québec usage differs for some of these marks, so match the audience rather than the speaker. If you edit cues by hand, the SRT generator page explains the format.

Septante, nonante and regional numbers

Belgian and Swiss French say septante (70) and nonante (90), and parts of Switzerland say huitante (80), where France uses soixante-dix, quatre-vingt-dix and quatre-vingts. If figures come back as digits this doesn't matter; if they come back as words, confirm the value and whether you want the speaker's form. Long compounds such as quatre-vingt-dix-sept are worth a second listen in prices and dates. Speakers from West and Central Africa often use local place names, personal names and loanwords that the transcript will spell by sound, so ask for a list of names before you review.

Casual speech on the page

Spoken French drops sounds freely: chais pas for je ne sais pas, y a for il y a, t'as for tu as. See whether your transcript writes these colloquially or restores the full form; either can be right. Subtitles usually read better with the full form, while research on how people actually speak should keep what was said. How to proofread an AI transcript gives an order of work.

Anglicisms and acronyms in French business speech

French meetings borrow heavily from English and then conjugate the borrowings: faire un call, checker, forwarder, on a un meeting, c'est trop corporate. Spellings of these hybrids vary even among native writers (checké or chéqué, mail or e-mail, ok or OK), so pick a house form and unify it. Acronyms are the opposite problem. French speakers spell many of them out letter by letter (PDG, RH, SNCF, TVA) and pronounce others as words (ONU, SMIC); check that each appears as the acronym your readers expect and not as a string of single letters or a phonetic guess. Company names pronounced the French way, such as Google or Uber, are worth one search each.

A Belgian webinar, start to finish

1. Export the 40-minute webinar recording from the meeting platform as MP4.
2. Transcribe with French selected, so nothing is translated: 40 credits (4¢).
3. Fix speaker and product names with find-and-replace, then run a French grammar checker for agreement.
4. Decide on typography (space before ?, guillemets, septante or soixante-dix) and apply it to the SRT.
5. If Dutch-speaking colleagues need it, run the file again with Dutch as the target for another 40 credits.

For recorded classes rather than webinars, see lecture transcription.

Using the tool on this page

• The spoken language is detected automatically; you select the output language — the spoken language itself for an untranslated transcript, or another language for a translation. Files can be up to 2 hours long.
• Only the audio is sent for processing — the video stays on your device — and audio and text are deleted within 30 minutes of delivery (how files are protected).
• You can download a timestamped transcript, subtitles (SRT and VTT) and plain text. Every output, step and limit is explained on the video to text page.