Library
Transcribe
Turn a recording you already have into a transcript, or open one you saved earlier. Live recording is on the Transcribe live tab.
Audio or video recording
Choose an audio or video file (mp3, m4a, wav, ogg, mp4, mov, webm and similar; only the sound is used). Either play it through the speech engine as if it were live (repeatable, good for comparing providers) or, with a provider that has a file service (Soniox, Modulate or the local engine), transcribe the whole file much faster than real time. It uses the provider chosen in Settings: .
A recording from a link
Paste the link to an audio or video file on the web (it should end in .mp3, .m4a, .wav or .mp4) and it is fetched here and run like a file you chose. It works only when that website lets other pages read its files; if it does not, download the recording and use the button above. Use recordings you are allowed to use.
A transcript file
Open a transcript saved by this tool (.json, .md, .srt or .vtt) to read or edit it. It opens in the editor.
Demo
See how it looks without a key or a microphone: a simulated phone call with two speakers, shown on the Transcribe live tab.
Transcription Configuration
Configure your API credentials and display settings. Keys are only ever kept in this browser - choose below how long and how safely.
Transcriber
The speech-to-text service, your key for it, and the language spoken. Choose a row; the chips show only the ones that fit, and the headings sort.
Your API keys are locked with a passphrase.
local-engine/token-helper.sh. It forwards your key to AssemblyAI for that one call and does not store or log it. Files do not need the helper; with it, finished file transcripts are also deleted from your AssemblyAI account.
Sources
Where the sound comes from when you press Record.
Saving
Where recordings are kept while you make them.
Board-part1.webm). Each bubble then gets a speaker button that plays its words. Make sure everyone in the room knows they are being recorded. Nothing is recorded while paused.
How long one file may run before the next part starts: 1h, 30m or 60s (or 1h30m). Without a unit the number is minutes. Between 10 seconds and 12 hours.
So the speaker buttons keep working when the transcript is opened later. Audio files can be large; they are stored in the same place as the transcript.
Notices
Telling people they are being transcribed: the Announce button on the Transcribe live tab speaks this out loud, and a line can show on screen while recording.
announcement.mp3 (or whatever type it is). A file with that name placed in your recordings folder is used too, when there is no copy in the browser. The volume below applies to a recording as well.
Privacy
What is kept, for how long, and how to clear it. Nothing here is sent anywhere: ZumiScribe has no server of its own.
Auto Correct
Words the transcriber keeps getting wrong, fixed as the text arrives.
.json file (or .csv, which opens in a spreadsheet) to back the list up or share it with a team. Import reads either, and also plain lines such as Creation, Croatia -> Creatio or heard as, write as.
Clean-up
What readers see, without changing what was said. The transcript keeps the real words: the saved file has them, they show while you edit a bubble, and switching these off brings them back. Copy, the exports (except JSON), the large captions, the subtitles window and the phones following along get the cleaned text. Changes here apply immediately.
* also covers longer words that start with it: fuck* masks fucked, fucker and fucking. Without the *, dick leaves Dickens alone.
Glossary
Explains initialisms and abbreviations in the transcript, and any word or phrase you right-click and choose "Look up".
{keywords} is replaced by the word, for example https://127.0.0.1:8765/abbr/{keywords}. It answers in the same JSON as the Free Dictionary, and its meanings may contain links ([Project page](https://intranet/pxp)) that open in a new tab. See GLOSSARY.md.
The Free Dictionary (freedictionaryapi.com, Wiktionary content), and for "Look up" also Wikipedia. Only the word or the selected text is sent, when you ask. Leave it off if the words must not leave your network.
Summary
A short summary, key points, decisions and action items, written on this computer.
OLLAMA_ORIGINS="https://zumilabs.com" ollama serve.
Services
Where the follow-along relay, the company glossary and the AssemblyAI helper are: in the ZumiScribe Local app on this computer, or on a company server. The speech engine and summaries have their own settings.
Sharing
Let people follow the live captions on their phones.
local-engine/follow.sh (see FOLLOW.md). Pressing the button starts broadcasting and shows a link and a QR code; the button is red while it is broadcasting. Only the caption text is sent, never the audio, and everyone with the link can read it as it is spoken.
Starts broadcasting when you start recording and stops when you stop (it needs sharing to be enabled).
Where the relay runs (this computer, by default).
For people outside this network: the https address a tunnel gives the relay (cloudflared tunnel --url http://localhost:8780 or ngrok http 8780). Never tunnel the speech engine, only the relay.
Phones are asked for it before they can read. It is added to the links and the QR code.
Send to a web address
Post the finished transcript to a web address you choose: a Slack incoming webhook, Zapier, Make, n8n, Discord, Teams or your own service. Nothing is sent unless you press Send or switch on automatic sending.
Display
How the transcript looks on screen.
Preview
Hello, my name is Alex. Can everyone see this from across the room?