Ovvos Wasel 1.0Two different lists: what Ovvos Wasel can hear (recognise as the source of speech) and what it can speak (deliver to listeners as interpreted audio).
Hears — 15 languages
| Language | Native name | Notes |
|---|---|---|
| English | English | |
| Arabic | العربية | Recognition is strongest on Modern Standard Arabic; Gulf, Egyptian, Levantine and Maghrebi dialects are recognised less accurately |
| Urdu | اردو | |
| Persian (Farsi) | فارسی | |
| Turkish | Türkçe | |
| French | Français | |
| German | Deutsch | |
| Spanish | Español | |
| Russian | Русский | |
| Italian | Italiano | |
| Chinese (Mandarin) | 中文 | Captions in simplified or traditional script (operator's choice) |
| Chinese (Cantonese) | 粵語 | |
| Hindi | हिन्दी | |
| Korean | 한국어 | |
| Japanese | 日本語 |
How the source language is handled
- Automatic detection (default): each sentence's language is detected, so speakers can switch languages mid-meeting.
- Expected languages: the operator can tick the languages they know will be spoken. This removes whole classes of misrecognition; a sentence detected in a language that is not on the list is dropped rather than forced into the wrong one.
- Fixed source: detection can be turned off and one language fixed.
Speaks — 10 languages, any 4 at once
| Language | Native name | Voice | Caption direction |
|---|---|---|---|
| Arabic (Modern Standard) | العربية | Neural voice by default; a standard voice as fallback | Right to left |
| English | English | Standard voice; neural voice selectable | Left to right |
| German | Deutsch | Standard voice; neural voice selectable | Left to right |
| French | Français | Standard voice | Left to right |
| Spanish | Español | Standard voice | Left to right |
| Italian | Italiano | Standard voice | Left to right |
| Russian | Русский | Standard voice | Left to right |
| Turkish | Türkçe | Standard voice | Left to right |
| Persian (Farsi) | فارسی | Standard voice | Right to left |
| Urdu | اردو | Standard voice | Right to left |
"Neural" voices are the more natural-sounding option; "standard" voices are lighter voices — clear and intelligible, recognisably synthetic. The operator chooses per language under Settings → Models → Voices.
Why four at once: every active output language costs one translation and one voice on every sentence. Four keeps the room close to live on a single PC. The limit is enforced everywhere languages are chosen, with an explanation.
Why Chinese is heard but not spoken: no on-premise voice is offered for Mandarin or Cantonese yet, so neither is offered as an output — rather than offering it and producing silence. Speak Mandarin or Cantonese and the room hears it in their own languages. (The optional cloud profile can speak them.)
Arabic
- Output is Modern Standard Arabic, understood by educated speakers across dialects.
- Arabic is spoken slightly faster by default. Arabic renderings run a little longer than the original, and this keeps listeners close to live.
- Arabic, Urdu and Persian captions render right to left, in Noto Sans Arabic, on phones, tablets and the room display.
Glossary
A per-language glossary fixes names, ranks, units and designations so they are never paraphrased and never drift between meetings. One line per term:
Northern Line = ar:الخط الشمالي, fr:Ligne Nord, de:Nordlinie
# lines starting with # are notes