C. FavreFrançais00:02
On doit valider le calendrier avec la commune avant jeudi.
Multilingual meetings
One meeting, several languages, one transcript.
LocalTranscript transcribes meetings in which people speak German, French, Italian or English, separates the speakers and labels the language of every segment. All of it on your computer.
Three separate jobs, done separately.
A useful multilingual transcript answers three different questions:
Speaker diarization
Who spoke when. Voices are separated into segments and labelled locally; name a speaker once and the app recognises them in later meetings.
Language identification
Which language each segment is in. The label sits next to the speaker, so you can see where a meeting switches from French to English.
Transcription
What was said, written down in the language it was said in.
What a mixed meeting looks like.
Four people, four languages, one timeline. Each segment keeps its speaker, its timestamps and its language label.
This is not translation: the French stays French and the German stays German. The transcript records the meeting as it happened.
Which languages?
German, French, Italian and English are the combinations Swiss teams mix most, and the ones shown here. Transcription uses multilingual Whisper models, so many other languages they support work the same way.
Swiss German dialect is the exception: it has its own dedicated model and is not part of the automatic multilingual mode.
Speed or accuracy.
Two kinds of multilingual models are available:
- Everyday models: fast, light on hardware, fine for daily meetings and dictation.
- Whisper Large (Pro): the most accurate option for supported languages; a bigger download, more memory and longer processing.
What to expect in practice.
- Language labels apply per segment: a clean transcript needs reasonably orderly turn-taking. Heavy crosstalk is hard for any diarization system.
- Short interjections in another language may be attributed to the segment around them.
- Nothing is translated: the transcript preserves the spoken languages.
Common questions.
Does it translate the meeting into one language?
No. Transcription preserves the language that was spoken, segment by segment. The language labels make the switches easy to follow.
How does it handle people talking over each other?
Speaker separation works best when people speak one at a time. Sustained crosstalk degrades every diarization system, including this one; expect to tidy those passages by hand.
Is Swiss German part of the multilingual mode?
No. Swiss German has a dedicated model and is selected separately. The multilingual mode is for meetings that mix Standard German, French, Italian and English.
Does any of this need a cloud service?
No. Diarization, language identification and transcription run on your computer. A cloud provider is only involved if you configure one yourself.
Run your next mixed meeting through it.
Record or import a multilingual meeting and read the transcript it produces. The free version needs no account.