Why Google Meet and Teams captions listen for one language at a time

Google Meet and Microsoft Teams captions turn speech into text against one spoken language that has been chosen for the meeting. Speech in any other language is decoded as if it were the chosen one, which is why a meeting that mixes two languages produces captions that read fluently and say the wrong thing. Both platforms now notice some mismatches, and Teams offers a language per participant with an add-on licence, but each speaker is still heard in one language. This guide is for anyone whose meetings on Google Meet or Microsoft Teams mix languages, and for anyone wondering why a notetaker built on captions inherits the problem. It describes what Google's and Microsoft's own help pages say as of October 2026, then how TrackHeros works around it with automatic language detection.

How does Google Meet decide which language to caption?

From one setting, the language of the meeting. Google's page on translated captions in Google Meet gives the path — in the meeting, More options, Settings, Captions, then "Select Language of the meeting" — and translated captions then translate from that language into the one you prefer. Google's page on live captions in Google Meet lists the caption languages, including Arabic for the Emirates, Egypt, the Levant and the Maghreb, all marked beta, and notes that "the captions are turned on only for you".

Neither page describes a way to caption two spoken languages at once. You choose one, and translation works from that one.

What changed this year is detection. Google's Workspace Updates announcement of 26 March 2026 says Meet now "detects the language being spoken during a session one time per-meeting". If the detected language does not match the meeting's language setting, a prompt appears after about 30 seconds of continued speech in the other language, with Change language or Ignore. The announcement says it is for meeting transcripts, summarised notes and recorded captions, and lists English, French, German, Italian, Japanese, Korean, Portuguese and Spanish. It corrects a wrong setting. It does not make a meeting bilingual.

How does Microsoft Teams decide which language to caption?

From the meeting spoken language. Microsoft's page on live captions in Teams meetings says to "set a meeting's spoken language to the language people are speaking for more accurate captions", chosen from a Meeting spoken language dropdown where you "select the language everyone is speaking". If people speak a different language, Teams detects the mismatch and prompts the organiser, a co-organiser or whoever started transcription to update it, with a "Yes, change it" option. Other participants can change it only after transcription has been turned off. Arabic (Saudi Arabia) and Arabic (United Arab Emirates) are in the list of supported spoken languages.

Teams goes further than one language per meeting for those with an eligible licence. Microsoft's page on multilingual speech recognition says that when it is turned on, "each participant can select their own preferred spoken language". It is turned on in the meeting's options; it needs an add-on licence, Teams Premium being one; and its supported languages are English, Spanish, French, German, Italian, Portuguese, Chinese, Japanese and Korean. Arabic is not on that list. The translated captions and transcripts it produces "are only available during the meeting".

So in Teams the unit can be the participant rather than the meeting. It is still one language per person.

Why does one language at a time break a bilingual meeting?

Because a recogniser decodes sound against the vocabulary of the language it expects. Speech in another language does not come out as gaps; it comes out as the expected language's nearest-sounding words, punctuated with confidence. And live captions have to produce text as the words arrive, so they cannot wait for the end of a sentence to discover that it changed language halfway.

Real meetings change language all the time. In a working meeting in Montreal, Mumbai or Beirut, the frame of a sentence is often French, Hindi or Arabic and the technical nouns English. A client's staff may speak their own language to each other and English to you. A prompt that offers to change the meeting's language helps when the setting was simply wrong. It cannot help a meeting that is genuinely in two languages, because whichever one is chosen, the other is decoded as if it were the first.

Per-participant languages narrow the problem without removing it. They help when each person sticks to one language. They do not help the person who starts a sentence in Spanish and finishes it in English, which is how bilingual people talk. The guide to meetings that switch language mid-sentence goes through what that does to a transcript.

Why does a notetaker that relies on captions inherit the problem?

Because the captions are its transcript. Many meeting tools collect the platform's captions during the call and summarise them afterwards. That is cheap and fast, and the captions carry something valuable: the platform knows whose microphone is live, so each line arrives with the speaker's real name.

But the words are only as good as the caption language. A summary of mis-decoded captions reads as fluently as a summary of good ones, and nothing in it says which half of the meeting was heard in the wrong language. Even where the platform notices a mismatch, the fix is a setting a person has to change during the call, and a meeting in two languages has no right answer to that setting.

The useful split is to treat the captions as a record of who spoke and when, and to get the words from somewhere that can hear more than one language.

How does TrackHeros use the captions and the recording?

TrackHeros treats the captions as the first pass and the recording as the second. The captions supply the real speaker names and the timing; the recording supplies the words whenever a meeting is in a language other than English or switches between languages.

	First pass: captions	Second pass: recording
Source	The platform's live captions, collected in the call	The recording, after the call
Language	One at a time, as the platform decodes it	Worked out from the audio itself
Contributes	Speaker names and timing	The words, rendered into English line by line
Used for the words	Only when the meeting is confirmed to be in English	When it is in another language, or in more than one

Whether a meeting needs the second pass is decided from the meeting itself, cheapest check first: a recurring meeting's memory of its language, whether the captions read as coherent text, a short audio sample, then a probe and a pass over the recording. Any of them can send the meeting to the second pass; none can keep it on the captions alone without independent confirmation.

A meeting in another language, or in two, is transcribed by a Whisper model that hears the spoken language itself. Each line is then translated into English on its own, keeping its speaker and timestamp, and a line that fails to translate keeps its original text rather than disappearing. The notes are written in English or in the language you chose, and the transcript page says, in one line, "Translated to English from a conversation held in French". A line that cannot be attributed to a speaker is labelled "Unknown" rather than guessed.

The decision is the same on Google Meet and Microsoft Teams, and there is no caption language for you to set. Close to 100 spoken languages are recognised, and the languages page lists them.