Transcription
Minutist turns the recording into text on your own machine. A live transcript streams while you record; a full pass runs when you stop. No audio leaves the device.
The models
Section titled “The models”Transcription uses NVIDIA Parakeet TDT (through sherpa-onnx) for English and the major European languages, with per-word timestamps. A Qwen3-ASR model provides a CPU fallback for broader language coverage. Minutist picks the engine from your transcription language; both run locally. See Models and hardware for sizes and requirements.
Set the language under Settings → Capture (the language picker). The default is English (auto-detection can misfire on short or quiet openings).
The transcript pane
Section titled “The transcript pane”Turn on the Transcript column from the pane toggle. It is a scrolling list of segments, each led by a timestamp. While you record, an audio-level meter sits at the top and new segments append live.
Each segment is tied to your notes and to the audio:
- Click a line: the notes scroll to the paragraph you wrote around that moment.
- Hover a note: the segments that paragraph spans are highlighted here.
- Drag the timestamp into your notes: it drops in as a quoted transcript chip.
- Play the audio: on a saved meeting, hover a line and click the play glyph to hear the original audio for just that segment.
Reprocessing
Section titled “Reprocessing”On a saved meeting (while nothing is recording), the Reprocess button in the transcript toolbar re-runs speech recognition and speaker identification on the recording, using the transcription language and model currently set in Settings. It runs in two phases (transcribe, then identify speakers) and replaces the stored transcript and speaker labels.
Translation
Section titled “Translation”Pick a language from Translate to… in the transcript toolbar and press Translate to render a translated view of the transcript, produced by the local model. The verbatim transcript is kept; Show original returns to it at any time. Translation covers a set of common languages; the picker lists them.