The complete guide for the person who runs Ovvos Wasel in a meeting room. For the short version, see the Quick Start Guide.
1. The three things you work with
| Where | What it is for | |
|---|---|---|
| The Ovvos Wasel window | The room PC | Setting up, starting and ending meetings, watching how it is going |
| The room display | The room's wall screen or projector | Showing participants how to join |
| Participants' phones | In their hands | Listening — they need nothing from you once they have joined |
The window can be closed at any time: Ovvos Wasel keeps running in the system tray (bottom right of the taskbar). The tray icon is green when audio is flowing and engines are ready, amber while starting, red when something needs attention. Its menu has: Open control panel · Show room display · Open listener page · Copy listener link · Quit.
Only one copy runs at a time. Opening it again brings the existing window forward.
2. Starting up
When you open Ovvos Wasel it loads three engines — speech to text, translator and voices — and shows each one's progress. This takes about 45 seconds.
Then it asks you to set the languages:
- Source — what they speak. Auto-detect is on by default and recognises any of the 15 supported languages. Ticking the languages you expect improves accuracy: a sentence detected in a language you did not tick is dropped rather than mistranslated.
- Target — what they hear. Up to four of the ten languages. The counter shows "n of 4"; choosing a fifth explains why it is not allowed.
From here, Start meeting goes straight to a live meeting; Setup opens the main window.
3. The main window
Top bar: profile (on-premise or cloud), capture state, listener count, and the pages Home · Live · Languages · Audio, plus Settings.
Home — "Ready for the meeting"
- Readiness dots (top right) — seven checks. All green means ready. Hover any dot for what it checks, what is wrong, and the fix.
- Hearing the meeting audio
- Sound is reaching the interpreter
- Detecting when someone speaks
- Recognition, translation and voices loaded
- Translations coming back
- Languages to listen in
- Phones on the WiFi can connect
- Source and target languages — tick and untick directly. Changes apply immediately, and show on every page that lists languages.
- Start meeting / End meeting, Open room display, Send test phrase.
- Who is listening — listeners per language, and the address participants use.
- Running a meeting — the three steps, as a reminder.
Live — during the meeting
- Latency — median and 90th percentile, end to end. Two to four seconds is normal.
- Listeners and sentences delivered.
- Per-stage timings — recognition, translation, synthesis.
- Losses — four causes, four fixes. All should stay at or near zero:
- Dropped — the system fell behind and skipped work (too many languages for the PC, or very fast speech).
- Not transcribed — sound was heard but nothing recognisable (noise, music, a language not expected).
- Not translated — translation came back empty.
- Never spoken — a translation could not be voiced.
- Counters restart at every Start meeting, so they describe this meeting.
Languages
The same choices as the start screen, plus: Only expect certain languages, a fixed source language (with auto-detect off), and the script for Chinese captions. Every change applies immediately.
Audio
What Ovvos Wasel listens to:
- Everything this PC plays — the usual choice, for a call on this PC.
- A microphone or line-in — in-room microphones or a mixer feed.
- One app only — just the conferencing application.
Plus the device, input gain, and a live level meter. Save and restart capture applies a change.
Settings
| Page | Use it to |
|---|---|
| Profile | Switch between on-premise and cloud (needs an engine reload) |
| Models | Replace the translation model; choose the voice per language; see installed language packs |
| Glossary | Fix names and terms per language — one line each: Term = ar:…, fr:… |
| Cloud keys | Enter providers and keys for the cloud profile (not needed on-premise) |
| Access | Set or randomise the listener PIN; change the administrator password |
| Privacy | Transcript retention, export, and text in logs — all off by default |
| Tuning | Sentence segmentation; engine parameters (advanced) |
| Diagnostics | Send a test phrase; test a translation into every target |
| Licences | Terms of use and acknowledgements |
Some settings take effect only after an engine reload — the window says so, names them, and offers Reload engines. A reload takes about a minute, and the room hears nothing while it runs: do it before the meeting, never during.
4. Running a meeting
The day before, or that morning
- Open Ovvos Wasel and wait for every readiness dot to turn green.
- Set the languages for this meeting.
- If you changed anything that needs a reload, reload now.
- Optional: on Access, press New random PIN for this meeting.
As the room fills
- Press Start meeting. Nothing is interpreted until you do.
- Open the room display on the wall screen.
- People scan, enter the PIN, pick a language. If someone joins while the engines are still loading, their phone says so and connects by itself.
During
- Nothing needs doing. If you watch anything, watch the Live page's loss counters.
- A participant whose phone says "Interpreter problem" is seeing the same issue your readiness dots show.
Afterwards
- Press End meeting. Interpretation stops on every phone immediately; anything not yet spoken is discarded; the room display clears its transcript. The counters stay, so you can review the meeting.
5. The room display
- Join card: QR code, address, PIN, languages. Light (yellow) or dark.
- Show full transcript: shows the original and a translation side by side, with times, from the moment you turn it on. Choose which translation appears. Off by default every time the display opens; it clears when the meeting ends.
- Open it from Open room display (Home, the meeting live view or the tray) — that way it carries the PIN it needs. Typing the address by hand on another machine needs
?pin=and the PIN at the end.
6. Good habits
- Good audio in beats everything. A line feed from the room mixer gives the best recognition; a laptop microphone at the far end of the table gives the worst.
- Tick the languages you expect. It removes a whole class of errors.
- Keep the glossary current with names, ranks, places and acronyms.
- Four languages is the ceiling for a reason. Each language costs a translation and a voice for every sentence.
- Tell participants that interpretation is machine-generated and for comprehension, not a binding record.
Ovvos Wasel 1.0