Short answer first: untick English only (better accuracy) on the Enter a stream URL page before you start, or click the EN badge in the transcript header to flip it to Multi while the stream is already running. Both send the same audio to a multilingual model instead of the English-only one. Here is what that looks like on a Spanish press conference.
Key takeaways
- English-only is the default because it is the most accurate model for English audio - not because other languages are unsupported.
- The EN / Multi badge in the transcript header is a button. It switches language mid-stream; captured lines stay as they are.
- On YouTube and Twitch each language is its own track off one audio pull, so two people can watch the same broadcast in different languages at the same time.
- Search and keyword alerts run on what was actually transcribed, so alerts can watch for a non-English phrase.
- Two tracks on one stream consume minutes twice, and changing language is a Pro feature.
What an English-only model does to Spanish audio
This is the part worth understanding, because the failure is silent. A speech model that has been trained to output English can only output English. Point it at Spanish and it does not report a problem - it writes down the closest English-sounding words it can find. A sentence about a central bank meeting starting at nine comes back as a string of English fragments that rhyme with it, and nothing in the interface looks broken.
Which means every downstream thing you wanted the transcript for quietly fails too. Searching for the phrase finds nothing, because the phrase was never written. A keyword alert on the word that matters never fires, because that word is not in the text. An export hands you a document that looks like a transcript and is not one. The clip above shows the two versions side by side for exactly this reason: the difference between them is not polish, it is whether the transcript contains what was said.
Before you start: the English-only checkbox
On the Enter a stream URL page, under the URL box, there is one checkbox: English only (better accuracy). It is ticked by default. Leave it ticked for an English broadcast and untick it for anything else - a non-English stream, or one that mixes languages - and the transcript opens in multilingual mode.
The default is not laziness. An English-only model is choosing between English words, and a multilingual model is choosing across many languages at once, which costs it some accuracy in each one. So the honest rule is: English audio, keep the box ticked; anything else, untick it. There is no upside to running multilingual mode on an English stream.
Mid-stream: the Multi button
The checkbox only helps if you knew the language before the transcript existed, and often you don’t - you open a live YouTube stream and the first speaker turns out not to be speaking English. So the language is also a control inside the running transcript: the small EN badge in the header is a button, and clicking it switches to Multi immediately.
Two things about that switch are deliberate:
- Nothing already captured is thrown away. Lines transcribed before the switch belong to the language they were transcribed in. The view reloads the history for the language you moved to, and new speech arrives in that language from that moment on.
- Nothing switches by itself. If the stream’s title is in a non-Latin script while you are still on English-only, the badge turns amber and offers a Non-English stream? Switch to Multi prompt. It is a nudge. The transcript never changes language on its own, because a language change part-way through is something you should be able to see in the record.
Two languages, one stream, at the same time
For YouTube and Twitch streams, switching language is not a global setting on the broadcast - it is which track you are attached to. Each language is a separate transcription track fed by the same audio pull, so when you click Multi the server attaches you to the multilingual track, starting one if nobody was on it yet. Anyone still on English keeps their English transcript running, uninterrupted.
That is genuinely useful on a bilingual event. One desk can follow the English track while another follows the multilingual one, from one broadcast, with both transcripts live and searchable. The cost is the obvious one and worth saying plainly: two tracks are two transcription streams, so they draw from your minute pool twice as fast as one. The second track exists only while somebody is actually on it.
The same control in the extension and the API
The browser extension carries the same badge in its side panel, and it remembers your choice between sessions - so if you mostly capture non-English audio from sites that aren’t YouTube, set it to Multi once and it stays there. Changing it in the panel applies to the capture you are running.
If you are driving LiveScript programmatically, the live subscription message takes a language field of "en" or "multi" and defaults to English - see the API documentation. Same two modes, no separate multilingual endpoint to learn.
When to use which
- English broadcast: keep English only ticked. It is the most accurate option for English and there is nothing to gain by switching.
- Entirely non-English broadcast: untick it, or hit Multi as soon as you hear the first sentence.
- Mixed or code-switching audio - an English presentation taking questions in another language, a bilingual panel, a press conference with translated segments: Multi, because that is the case the multilingual model is built for.
- You don’t know yet: start on the default and watch the first few lines. If they read like nonsense that half-rhymes with the speech, that is the signature of an English-only model on non-English audio - click the badge.
Try it on a stream that isn’t in English
Paste a non-English live stream into LiveScript, untick English only, and watch the words arrive in the language they were spoken in - then set an alert on a phrase in that language and see it fire. If you are new to live transcripts generally, start with getting a transcript of a stream that is still live, and if you keep missing the start of broadcasts, see streams that start without warning.