What's new in Jot
Three things that make everyday dictation better on Windows. Everything runs on your PC; nothing is sent anywhere.
Vocabulary that gets your words right
Add the names, products and jargon you use under Vocabulary in the left pane, and Jot spells them your way from the next dictation. No switch to turn on, nothing extra to download.
- Right the first time in English. The speech model listens for your terms while it transcribes, so a name like "Sriram" comes out correctly instead of being fixed afterwards.
- It never breaks what was already right. After each dictation a spelling check and a safety gate make sure a term only replaces a word that really was a miss.
- You can see and correct it. A small mark on the pill shows when a term was used, and in the recording you can right-click a word to pick the one you meant.
- 18 more languages get spelling fixes for near-misses, including French, German, Spanish, Italian, Dutch and Portuguese.
A real progress bar while Jot works
When you stop a long dictation, the pill shows "Transcribing… 42%" with a bar that tracks the actual work, not a spinner. You always know how long is left.
- The same style shows up when Jot finds speakers in a recording: "Finding words… 25%", then "Finding speakers…".
- A soft chime plays when a long transcription finishes, only if it took more than 5 seconds. Short dictations stay quiet.
- Long recordings and imported files of any length now transcribe. A one-hour meeting works just like a one-minute note.
Progress while Jot finds the speakers in a recording.
Who said what: speaker labels
Open a recording, choose Detect speakers from the "…" menu, and Jot splits the transcript into turns, one colour per person. It uses NVIDIA's Nemotron 3 Diarization model and handles up to 8 speakers.
- See it on the timeline. The playback bar shows who talks when, in each speaker's colour. Click anywhere to jump there.
- Name people once. Type a name in a speaker's box and every one of their turns updates.
- Edits don't break it. Fix a word in the transcript and the labels stay lined up.
- Fast and accurate. A 17-minute, 4-person meeting takes about 30 seconds on a laptop, with 99.9% of words assigned to the right person in testing.
- Subtitles too. Export as WebVTT for real timings and speaker names.
A two-person test conversation. The speakers have been renamed from "Speaker 1" and "Speaker 2".