Translated subtitles in your livestream, added as a Browser Source. If part of your audience follows in another language, this gives them the words while the speaker is still talking.
Tolka Live is real-time AI speech translation for live events; this page covers the
OBS side. The integration is one source. Every account has a permanent caption link —
https://tolkalive.se/overlay/?id=YOUR_ID — a web page that opens a WebSocket to
wss://rt.tolkalive.se, listens for the translation the app is publishing and draws it
bottom-centre on a transparent background. Words appear one by one while the sentence is still being
spoken, typically about a second behind the speaker.
Worth saying once, because it is the real difference from most other ways of getting translated captions into OBS: no plugin, no speech model on the streaming machine, nothing to compile. The stream computer renders a web page; the translating happens elsewhere.
Tolka Live&text=Test caption on the end: the page shows that line and never connects, so no
session runs and no hours are used.&text= off
the URL — while it is there you see the test line, not the live feed.The look is set by query parameters on the URL — you edit it in the Browser Source properties and press OK.
| Parameter | What it does | Example |
|---|---|---|
text | A fixed test line instead of a connection. For previews. | &text=Test caption |
size | Text size, percent of source height. Default 5. | &size=4 |
color | Text colour, hex without the #. Default white. | &color=ffffff |
bg | The box behind the text, as hex — or none for no box. | &bg=none |
bgopacity | How opaque that box is, 0 to 1. Default 0.45. | &bgopacity=0.25 |
bottom | Distance from the bottom edge, percent of source height. Default 6. | &bottom=14 |
lang | Which target language this source follows. | &lang=es |
They combine in any order:
https://tolkalive.se/overlay/?id=YOUR_ID&size=4&bg=none&bottom=14.
The page is transparent by default — that is what lets it composite over a video scene instead of covering it — and the text sits in a black box at 45 % opacity with a drop shadow. The box is the safe default: it reads over anything, a white shirt, a sunlit window, a slide with a photograph on it.
bg=none removes the box and leaves bare text with its shadow. It looks better over a
dark stage or graded footage, and it is also the first thing to fail over bright video. If the picture
changes a lot during a broadcast, keep the box and take the weight out of it instead:
bgopacity=0.25 separates enough without a black bar across the frame. If your scene already
has a lower third or a logo bug down there, lift the captions with bottom.
Up to five target languages are translated in parallel from a single microphone, out of 75+
available, and the source language is detected automatically. In OBS that is one Browser Source per
language: the same overlay id plus &lang= and the language code, for every language
other than your primary one.
Stack them in one scene at different heights — Spanish at &bottom=6, Ukrainian at
&bottom=16 — so every viewer sees both rows, or build one scene per language and send
each to its own destination. Viewers cost nothing either way, and the same translation reaches the
audience's phones by QR code at the same time. Watch the hours: translated time counts per language, so
an hour of speech in two languages spends two plan hours.
Captions are not the only output. Tolka Live can also speak the translation in an AI voice and send it to any output device — including a virtual audio device, the kind of loopback driver streamers already run on macOS and Windows. Point the voice at one, add it in OBS as an Audio Input Capture, and you have a translated audio track for its own output track. The voice follows a moment behind the captions.
&text=Hello and open the URL in a
browser. Text there but not in OBS is a scene problem; no text in either is a link or network
problem.wss://rt.tolkalive.se, and some venue networks block it. It retries on its own, backing
off up to fifteen seconds; a phone hotspot proves whether the network is at fault.Every plan includes every output — projector, ProPresenter, this overlay, the audience's phones and the AI voice — and plans differ only in translated hours per month. A one-off broadcast is usually the Event licence: 449 kr / €42 once, ten hours, valid twelve months. Regular streams fit the monthly plans, from 595 kr / €55 for fifteen hours a month, half price the first month. Prices exclude VAT; the full table is on the pricing page.
There is a free 15-minute trial first, no credit card: create an account in the portal. The step-by-step guide covers audio, ProPresenter and OBS in order, and the FAQ answers most of the rest.
No. The overlay is an ordinary web page added as a Browser Source, so nothing is installed on the streaming computer and no model runs on its GPU. The translating happens in the cloud and arrives over a WebSocket.
Yes. The caption link is hosted on tolkalive.se, so an operator somewhere else can add the same overlay. Both machines need internet, but not the same network.
Yes. Up to five target languages run in parallel from one microphone. Add one Browser Source per
language — same overlay id, different lang parameter — with different bottom
values so the rows do not overlap.
Add bg=none to the URL. The backing box disappears and only the text composites over
your video; the text keeps its drop shadow, so it stays readable on mid-tone footage.
They are good, but AI translation is not perfect — names, numbers and unusual expressions can come out wrong, especially with unclear audio. Treat them as support for viewers rather than certified interpretation, and use professional interpreters where accuracy is legally critical.
No. The audio is streamed to Google Gemini, which translates it in real time, and is not recorded or stored by Tolka Live. Gemini is a US processor.