Live transcripts of your Google Meet calls in Croatian and English, with who said what and your notes alongside. When the call ends, dinome polishes the transcript and writes the summary. It all runs on your own computer, so nothing is uploaded to be transcribed.
Each sentence appears a moment after it is spoken, with who said it. Grey text is a live preview that settles when the sentence ends. When the call is over, the summary writes itself, on your own graphics card.
A few sentences, then key points, decisions, action items with who and by when, and open questions, from the polished transcript. Written by a model on your own graphics card (Gemma 4 or Qwen 3.5), or by ChatGPT or Claude with your own API key.
Names come from Meet itself, your own voice from your microphone. When someone shares their screen and Meet stops showing who speaks, dinome recognises the voices, during the call and once more when it ends.
Speech recognition runs on your NVIDIA card, much faster than people talk. Nothing to upload, nothing to wait for.
Models fine-tuned for Croatian, with punctuation and capitals, and Whisper for English.
After the call, the transcript is corrected line by line: grammar and case endings, gender forms (“bio sam” or “bila sam”, from how each person speaks), names and misheard words, punctuation, with the filler words left out. The transcript as heard stays one click away.
A client calling from the car? After the call, dinome takes the road, wind and street noise out of the others’ voices, on your computer.
Click a line to hear it. The seek bar is the recording's waveform in each speaker's colour; hover it to see who said what. Clean plays the others with the noise taken out.
Search ignores capitals and Croatian diacritics. Several words find lines with all of them.
The calendar puts every meeting on its day, recurring ones in the same colour, busier days a little brighter. Click one to open it.
Jump to any meeting, or any action, by typing a few letters. ? shows every shortcut.
Connect Google Drive and every meeting is copied there, filed by date, as one readable document. Share it by link or invite people. Older meetings' audio can live in Drive only, to save space here; it comes back when you play it.
dinome follows your Windows theme, or your choice. Drag to compare.
Run the installer for Windows. Its setup guide downloads the speech engine and the Croatian model for you, with progress, and a short tour shows you around.
It comes with dinome: the setup guide opens its folder, you load it in Chrome once, and it waits in the toolbar.
dinome asks whether to take notes, and in which language: ask everyone first. The transcript appears as people talk, and is polished and summarized after the call.
dinome has no servers and needs no account. Your meetings are on your computer, and leave it only when you choose.
64-bit. Installs for your user, no administrator needed.
Recommended. Without one it runs on the processor, more slowly.
Version 116 or newer, for the extension.
Installed, with the Croatian model; 8 GB free during setup.
| Speech model | Word errors | Speed | Writes |
|---|---|---|---|
| Croatian · ParlaSpeech turboDEFAULT | 8.3% | 0.04× real time | Punctuation and capitals |
| Croatian · BuzzASR large-v3 | 10.2% | 0.12× real time | Lowercase, no punctuation |
| English · Whisper large-v3-turbo | – | 0.02× real time | Punctuation and capitals |
Measured on 40 Croatian FLEURS test sentences on an RTX 5070 Laptop GPU. Your meetings will differ from read news sentences.
Not to be transcribed: the speech recognition runs on your own computer, and the recording is kept in the meeting's folder there. If you turn on the Google Drive backup, a copy goes to your own Google Drive, and you can keep older meetings' audio only there to save space.
Croatian, with two models fine-tuned from Whisper large-v3, and English, with Whisper large-v3-turbo. Choose the language when the call starts, on the card dinome shows in Meet; each model understands one language.
An NVIDIA graphics card with 6 GB or more makes it fast. Without one, dinome runs on the processor, much more slowly. Summaries on your computer can also use AMD and Intel graphics.
You choose: a model on your own computer (Gemma 4 or Qwen 3.5, downloaded once), or ChatGPT or Claude with your own API key, which also fix more of the words speech recognition gets wrong. With an online model, the transcript (and your notes, for a summary) is sent to it, never the audio.
After each call dinome polishes the transcript: grammar, case endings, gender forms, names and misheard words, punctuation and filler words. Each correction is checked, so one that would change what someone said is left out, and the transcript as heard stays one click away.
After the call dinome cleans their voices: it takes road, wind and street noise out of the others' audio, on your computer, and Clean in the player plays it that way. The recording itself is kept as it is.
Today dinome works with Google Meet in Google Chrome.
dinome does not join the call or appear in it: it listens to what your Chrome tab plays and to your microphone. So tell everyone, and record only with their consent: dinome asks you to agree to that before your first recording, as the law asks in Croatia and many other countries.
On your computer, in %APPDATA%\dinome\meetings, one folder per meeting. Settings → Storage moves them to another folder or drive, and
Settings → Backup copies them to Google Drive or a folder that OneDrive or Dropbox syncs.
It comes from the Latin dixi, notavi, memini: “I spoke, I noted, I remember”, after veni, vidi, vici. That is what dinome does with a meeting, and the first syllables of the three words, di · no · me, make the name.
The installer is not code-signed yet, so Windows SmartScreen asks once: click More info, then Run anyway.
Download dinome for Windows, add the Chrome extension, and join your next call.
Questions or problems: dinome.support@mariosem.es