Disconnected 0 listening
Audio & languages
Microphone
Speaking
Translate to
Microphone sensitivity
Silence below start streaming to measure
Leave the room quiet for a few seconds while streaming and watch quiet floor. Set the threshold just above it — the suggestion does this for you. If it is too low the recogniser hears the room and invents words; too high and a softly-spoken person gets cut off.
Words to listen for
Names (fi)
Names and words the recogniser keeps getting wrong — people, places, your organisation, terms that come up every week.
Separate them with spaces, not commas. If you do type a comma it is turned into a space when you leave the box, so you never need one. Names made of several words are fine — just write them normally, with a space between.
Gives the recogniser memory across sentences, which helps it keep names and spellings consistent. The cost: during silence it can repeat something already said, because with no sound to work from it continues the text instead. Turn it off if you see whole earlier sentences reappearing. The word list above keeps working either way.
Keep the list short and put the most important words first: only the beginning of it reaches the recogniser, and a long list dilutes the effect. It nudges the recogniser toward these words, it does not force them. Saved separately for each spoken language — changing Speaking above swaps to that language's list.
Translation
Context previous sentences
How much of what was already said the translator may look back on. 3 is a good default; higher can make it repeat itself.
Reads the translation aloud to listeners who asked for audio.
Takes effect on the next sentence — safe to change mid-service.
Translation help
About this event
A sentence or two about the setting, in any language. The translator reads it to choose the right tone and vocabulary — a sermon, a lecture and a business meeting each want different wording. Leave it empty if you are unsure; nothing is added then.
Glossary (fi)
No glossary entries yet.
How specific terms should be translated. A row appears here automatically for every word you typed in Words to listen for above — fill in a meaning only where the obvious translation would be wrong, and leave the rest blank.
Only rows with a meaning are sent. A meaning is guidance for when that term actually comes up, not an instruction to use the word everywhere. Use it for proper nouns and in-house terms: Ratina → Ratina (a venue, do not translate). Like the word list, it is kept per spoken language.
Sentence handling
words
Breaks a run-on sentence at "and" / "but" once it passes this length, so listeners are not left waiting.
seconds
Watches the screen, not the microphone. If no translation has appeared for this long it sends whatever is pending, word for word, with no check that it forms a sentence. A blunt last resort — it cannot tell "the pipeline is stuck" from "the translator is slow", so it also fires during normal delays. Leave it off unless text is getting stuck.
seconds of mic silence
Watches the microphone. When you stop speaking for this long it sends the last unfinished phrase, so a short sentence at the end of a thought is not left hanging. This is the trigger the two settings below depend on. Uses the microphone sensitivity above to decide what counts as quiet.
Trim silence
Drops words the recogniser invents during a pause.
s
Only active while “Send tail after … seconds of mic silence” above is on.
Broadcast
(off)
Room code
Account
not signed in
Advanced — not part of the operator UI
Server
Translation
s
Alt in
Alt out
Speaker ID Not enrolled
Browser audio processing
Audio diagnostics
not streaming yet
Alt-profile hotwords
Alt-profile translation hints
Glossary (Alt)
No alt hotwords.
Live Translation
Sign in to broadcast from this device.
What language do you speak?
Enroll your voice
so the system knows who is the owner
Speak naturally for 5 seconds
0.0s 0ms -- 0 0 German