Recorder
The Recorder records a conversation held on this machine. It keeps two tracks: your microphone, and the sound coming out of the speakers. A call in Meet, WhatsApp, Telegram or Discord works the same way, and so does a video. The Recorder needs no plugin for any of them. It also imports audio files and keeps a library of transcripts.
Opening it
The Recorder is off the menu bar on a fresh install. Right-click an empty part of the menu bar and pick it from Add a widget. A meeting card on the board can also open it and start a recording linked to that appointment.
Recording
Record starts a capture. Stop ends it. While it runs, the window shows a red dot, a running clock, a level bar per track and a live voice wave.
The speaker icon next to the button is the second track. Its tooltip reads Also capture the speakers (the other side). The app shows this icon only on a machine that can share its audio output. A line under the controls says what is really captured:
- Ready, both sides. before you start, and Recording both sides. while it runs.
- System audio was refused: recording your microphone only.
- The system stream carried no audio: recording your microphone only.
- This system cannot share its audio output. Your microphone is recorded; the other side is not.
- On a Mac older than macOS 13, a line says that version is needed.
Your microphone is always recorded, even when the other side cannot be.
Two lines stay on screen. The first is always there: The others are recorded too, and nothing tells them. That is on you. The second appears when the speakers are captured: Without a headset, the speaker echo gets transcribed too.
The window cannot be closed by accident during a recording. Closing it asks A recording is running. with Keep recording and Stop and close.
The audio is written as you go, as 16 kHz mono WAV files. Recordings live in
~/.mnemosyne/recordings. An hour of audio takes about 115 MB.
Follow live
While a recording runs, Follow live loads the transcription model for the length of the meeting. A card titled During the meeting then appears on the board with the newest lines and the topics that come back. Its status says Live, how many seconds it is behind, or why it paused. Stop live view unloads it. The transcript you keep is still the one made after the recording.
Importing audio
Import offers Import files and Import a folder. You can also drop files or a folder anywhere on the library. The Recorder reads mp3, m4a, wav, flac, or the audio track of a video.
Each file is converted to the format the transcriber reads and copied into the library. One folder brings 500 files at most. An imported folder becomes a folder of the library.
The library
The library lists every recording, newest first. Each row shows one status pill: Recording, Transcribing, Queued, Failed, Transcribed, To transcribe, Interrupted or Unreadable. Interrupted means nobody pressed stop: the app quit or the window closed mid-recording.
Above the list sit a search field and filter chips: All, Transcribed, To transcribe and No audio. Each chip shows its count. The last chip you picked is remembered.
Select turns on a selection. The selection bar then transcribes, deletes the audio of, or deletes every selected row, and names the count. A row keeps its own actions behind its menu: Open, Rename, Open the folder, Save as Markdown, Delete the audio and Delete.
Both destructive actions ask twice, in place. The first press says what will happen, for example how much disk it frees. The second press acts. The armed state lapses after a few seconds.
Delete the audio removes the WAV files and keeps the transcript. The row then reads Audio deleted.
A recording is named by its date until you rename it. The name titles the transcript and the file it is filed under. Leaving the field saves the name. Escape cancels. An empty field gives the date back.
Folders
New folder creates a folder. Drag a recording onto a folder to move it. Move to does the same from a selection, with the keyboard. Deleting a folder keeps its recordings, outside any folder.
When the library is at least 820 pixels wide, it shows tiles, with the folders in a sidebar. A narrower window keeps one list, and each folder is a card in it.
A folder can file its transcripts into a vault. File into a vault offers the vaults that read Markdown, or New vault. Every transcript of the folder is then written there, including the ones already done. Stop filing this folder stops it. Transcripts already filed stay in the vault.
Transcribing
Transcription runs entirely on this machine, with the local Whisper engine. If the engine or a model is missing, the panel says so and points to Settings › Voice › Local transcription. See Voice settings.
Transcribe starts one recording. A folder offers to transcribe the rest of its files in one queue. The queue bar says which file it is on, out of how many, and how much audio is left. It shows no percentage and no time estimate. Stop halts the queue.
The panel shows the measured share of audio already heard, the elapsed time and, once there is a rate, the time left. The model you pick in Model is remembered for the next file.
Reading a transcript
Open a recording to read it as a conversation. Lines are grouped into turns. Each turn is labelled You or Them by its track. A one-track import shows Audio, or You when your enrolled voice matches.
- A search field filters the conversation.
- Play this passage plays the audio behind a line. Whisper mishears names and numbers, so you can check.
- A player bar plays the whole recording. The line being heard lights up.
- Copy the transcript and Save as Markdown export it.
- Transcribe again replaces the transcript.
Without a headset, the microphone also hears the speakers. The Recorder marks those lines on your track as speakers and folds them away. It never deletes them. A button shows them again.
After the meeting
Three buttons sit under a transcript. Each one waits for your press, and nothing is written until you confirm.
Summarise writes a note. Before it spends anything, the panel shows the speakers the model will read, an estimate of calls and characters, and a field labelled Context: who is talking, what this is about. A name in the note can only come from that field or from the recording. After the run, the panel shows the cost read from the app's own call journal. The note is kept for that recording, and File it puts it in a vault.
Tasks and appointments reads what the conversation decided. It opens three dialogs in turn: the tasks, the promises, and the follow-up appointments. You can edit or drop each item. Tasks and promises go to the To-do. Appointments go to the Agenda, and also to Google Calendar if you tick Also in Google Calendar.
Recap by mail drafts a mail to the participants of the linked appointment. The draft starts from the summary note, so it needs that note first. Send… lists the recipients, and only Confirm sending sends. Save to Drafts sends nothing.
The tasks analysis is kept per recording. A second press shows it again without a new model call, and Run again asks the model again.
My voice
The person icon opens My voice in the Voice page of Settings. The app learns your voice from 3 samples of 30 seconds, made on 2 different days. Calibrate my voice records one sample while you read a text aloud. The recording stays on this machine. The app then tells your voice apart on a one-track import.
From afar
Zoom the canvas out and the Recorder shows one state at a time. A running recording shows its clock and voice wave. A running transcription shows its progress ring. A playing recording shows its name and position. Otherwise it shows how many recordings there are and how many wait. A press brings the camera to the window.
From afar the Recorder also lists the Latest recordings, five at most, and the time transcribed this year. Each one has three doors: Summary, Tasks and appointments and Transcript.