Réunions

Meetings

Record your meetings and receive the transcript, with each speaker named.

In preparation Create an account

Réunions records from the browser, transcribes the discussion and tells who spoke: the minutes start from a text that is already written.

Réunions is in preparation. It will join the Prophetix subscription catalogue when it is released.

Every feature of Réunions

66 features in 12 areas

Captures your microphone and the system audio (what the other participants say) at the same time, with a single click on Start.
Separate level meters for the microphone and the system audio, to check that everything is being captured.
Meeting name entered in advance: it is used to name the folder and the files (date, time, name).
Option to reduce microphone echo while the system audio is playing (not needed with a headset).
Confirmation requested before starting a recording when a meeting is already displayed: a new timestamped folder is created, nothing is overwritten.
New meeting button to start again from a blank workspace, without erasing anything on the disk.

Detects the start of a Teams, Zoom or Webex meeting (Slack and browser-based video conferences included) and offers to record.
Never starts a recording by itself: you are always the one who decides.
Takes the meeting name from the application window when possible.
Detection can be turned off; manual recording is always available.

One timestamped folder per recording, stored by default in the Music folder, with a location you can change.
A light mixed MP3 (96 kbit/s by default, up to 320 kbit/s), ready for listening or for sending to other tools.
By default, two separate tracks as well: your microphone alone, and the participants alone.
Variant of the mixed file with the microphone on the left and the system audio on the right, to isolate each voice in other tools.
Transcript as readable text (TXT) and as structured JSON, with speaker, time boundaries, detected language and confidence.
The settings (model, bitrate, folder, languages, separate tracks, automatic playback) are set in a settings file.

Local transcription by the Whisper model, started on demand during review: Transcribe button.
Five models to choose from, from fastest to most accurate (Tiny, Base, Small by default, Medium, Large v3 Turbo), downloaded once and then kept on the computer.
Language of your choice among all those Whisper supports, including French, English and Arabic (Lebanese included), or automatic detection.
Meeting in two languages: automatic detection chooses the language for each slice of a few seconds.
Choice of the track to transcribe: microphone and participants separately, a single track, or the mixed file.
Two progress bars: decoding of the MP3, then generation of the transcript.
Can be regenerated at any time with another model or another language; the notes are kept.
Live transcription during recording, turned off by default because it is very demanding on the processor.
Filters against sentences invented from silence or background noise.

Tells “Me” and “Participants” apart without guessing: your microphone is you; the system audio is the others.
Optional separation of the participants from one another (“Participants · Speaker 2”), adjustable with the number of people expected.
Separation done on your computer, without Python: models of about 32 MB downloaded on first use, which can be deleted from the application.
A failure of speaker separation never makes the transcription fail.

Drop zone in the Meetings pane: drag an audio file or a video, or click to choose it.
Audio taken as is: MP3, M4A, WAV, AAC, WMA, FLAC.
Video (MP4, MOV, MKV, AVI, WMV, MPG, WebM and others): the audio track is extracted and converted to MP3, and the original is never touched.
An imported file becomes an ordinary meeting: playback, waveform, notes and transcription.
Formats that Windows cannot read are refused as soon as the file is hovered over; an unreadable file is never left half-imported.

List of all past meetings, stored in a local database on your computer.
Search by name, ignoring case and accents.
Sorting by newest or oldest date, by name or by duration.
A meeting can be removed from the list without deleting a single file: it does not come back on its own, and a marker in its folder lets you make it reappear.
Opening of the meeting folder in File Explorer.
Automatic recovery of the folders already present on the disk at startup.

Audio player with a clickable waveform: click or drag to move around in the recording.
Clicking a sentence of the transcript jumps to the moment it was spoken, with automatic playback that can be turned on.
The transcript scrolls to follow playback, and the current sentence is highlighted.
Correction of the text of a sentence directly in the transcript.
Notes timestamped at the moment being listened to or at the moment of the recording.
Search in the transcript with an occurrence counter and navigation from one occurrence to the next.
Separate, resizable transcript window, synchronized with the application.

Export of a meeting as a ZIP archive: audio files, transcript as text and as JSON, notes.
Copy of the whole transcript (timestamps and speakers included) to the clipboard.
The mixed MP3 can be imported as is into summarization tools such as Plaud or NotebookLM.

Windows 11, on an Intel/AMD (x64) or ARM64 processor; Windows 10 is not supported.
Recordings, transcripts and models stay on your computer; the network is used only to download the models and to sign in.
Sign-in with your Prophetix account; without a valid license, the application becomes read-only (review and export remain possible).
Self-contained package: no runtime to install, but the Visual C++ 2022 library is required.

Recording a meeting with other people often requires their consent: the application does not replace checking the applicable framework.
A headset is recommended: without one, the microphone also picks up the speakers and the participants' sentences are transcribed twice.
No graphics card acceleration on Windows ARM64: transcription runs on the processor and the largest model is slow.
Live transcription lags by 15 to 20 seconds.
Separation of the participants from one another is limited to three simultaneous voices per ten-second window.
Meeting detection has no effect with virtual microphones, and the name of a Google Meet meeting is not always found.
What exists today is the Windows desktop application (Transcribe): recording from the browser is not written yet.
The application is still in preparation: no web version, and no Linux or Android version.