Translate French Audio to English

Upload the French audio (MP3, WAV, M4A, OGG, FLAC) or video (MP4, MOV, WebM). Auto-detect confirms French (override manually for Quebecois, Belgian, Swiss, or African French edge cases). Set the target to English. Export as TXT, DOCX, SRT or VTT. Median processing time is 3-5 minutes per hour of audio, measured across 7,432 production jobs. First five minutes are free without an account; 30 minutes on signup, no credit card. Note the output is English text and subtitles, not a dubbed English voice.

Yes on standard broadcast Quebec French — Radio-Canada or TVA style speech behaves much like France French. Informal joual, the Montreal working-class register, is harder: different vocabulary and phonology, and less of it in the training data. We have not measured the gap, so we will not put a number on it. Practical rule: broadcast or educated conversational Quebecois is fine to work from; heavy joual is a first draft that needs a Quebecois reviewer.

Yes, with caveats. Ivorian, Senegalese, Moroccan, Congolese and Cameroonian French all transcribe, but these varieties are less represented in the training data than France French, so expect more errors. We have not benchmarked them and will not invent a figure. Usable for content research and first-draft translation; anything broadcast-grade needs a bilingual reviewer. Standard business or media speech works noticeably better than informal speech or a heavy regional accent.

We do not publish an accuracy percentage for this, and the reason is worth stating. The public benchmark people cite is OpenAI's Whisper paper (arXiv:2212.04356), which reports French at 8.3% word error rate on the FLEURS test set — worse than English (4.2), Spanish (3.0) and German (4.5). French is not the easy pair it is often assumed to be: liaison and a large set of homophones give speech recognition more to get wrong. But that figure benchmarks Whisper Large-v2 on clean read speech, and our standard tier runs Large-v3 Turbo while our premium tier runs a different engine again — so it does not describe our pipeline or your recording. Translation to English is then a second stage with its own error modes. The figure we do stand behind is speed: median 3-5 minutes per hour of audio, measured across 7,432 production jobs.

Yes. Upload MP4, MOV, WebM, MKV — VexaScribe extracts the audio track server-side, transcribes in French, translates to English. Export as SRT for subtitles on the original French video (drop into Premiere Pro, DaVinci, YouTube Studio). Timestamps preserved through translation.

Yes on the free tier — 30 minutes of transcription + translation on signup, no credit card. Translation itself is included at no extra cost on every plan. Premium tier (higher-accuracy internal model) available at 1 credit per 5,000 characters for client-facing content.

Yes. Reverse direction workflow: upload English audio, set source = English (auto-detects), target = French. Output uses standard France French. Common workflow: English podcast episode → French subtitles for France/Belgium/Africa distribution, or English brand content → French for Quebec/Canada localization.

WhatsApp exports voice notes as .opus or .m4a. Long-press the message → Share → Save to Files (iOS) or Save (Android). Upload the file to VexaScribe. Common LATAM-adjacent workflow: French-speaking African diaspora, French business contacts sending voice updates, French family communication.

France French (Parisian standard) — the strongest available, best training data coverage. Quebecois — solid on broadcast/educated speech, weaker on informal joual. Belgian French — standard coverage. Swiss French — standard coverage. African French (Ivorian, Senegalese, Moroccan, Congolese) — usable, but less represented in the training data, so expect more errors; we have not benchmarked the gap. Idiomatic French (verlan, argot, franglais slang) — machine translation handles these worse than formal speech.

Yes. Whisper Large-v3 diarization identifies distinct speakers (Speaker 1, Speaker 2, up to 50 speakers). Speaker labels persist through translation — Speaker 1's dialogue is attributed to Speaker 1 in the English output. Best on 2-6 clearly-separated speakers; heavy overlap drops attribution accuracy.

Yes. MP3 is the most-common upload format. Direct upload — no conversion needed. Same workflow as any other audio format. Common use case: French podcast episodes ripped as MP3 for translation to English show notes or subtitles.

Roughly 30 minutes end-to-end. Transcription is the bottleneck (roughly equal to audio length via Whisper Large-v3); translation stage runs nearly instant on the resulting text. Very short French clips (under 5 min) can finish in under a minute.