What happens if the notetaker gets the meeting language wrong?

TrackHeros decides each meeting's language on its own, so it can get one wrong, and the honest question is what that costs you. This answer is for freelancers and consultants whose Google Meet and Microsoft Teams calls are sometimes in a language other than English, or in two at once. The full picture of how detection works is on automatic language detection.

How does the notetaker decide the language?

Nothing has to be set. Each meeting's language is worked out from the meeting itself, cheapest check first:

1. A recurring meeting's memory of the language it has been held in.
2. A check of whether the platform captions read as coherent text.
3. A short audio sample, about two minutes, asked only "which languages are spoken here?".
4. A probe, and then a full audio pass over the recording.

The captions come from Google Meet or Teams and decode one language at a time. The audio pass listens to the recording itself and works out what was spoken, from close to 100 languages it recognises. More in how a notetaker can detect the language on its own.

Which way does a wrong guess fail?

Towards the audio pass. Every stage can send a meeting to the full pass over the recording; none can keep a meeting on the captions alone without a second, independent check agreeing. So the common mistake, sending a meeting to the audio pass that the captions could have handled, costs processing time and nothing else.

The harmful mistake is the opposite: trusting captions that were decoded in the wrong language. For that to happen, the captions also have to pass an independent check, so one fooled stage is not enough. Recurring meetings add one more guard. Only a series that has been heard in English alone can skip the audio pass; every tenth occurrence goes back through the checks rather than being taken from memory, and one sighting of a second language sends it back to the audio pass straight away. A series held in any other language gets the audio pass every time. How that memory works is in recurring meetings remember their language.

Can a setting make a wrong guess more likely?

One can. Listing the languages a client speaks is optional, and a language left off that list is treated as less likely: when English and an unlisted language are both heard, the unlisted one can be taken for a mishearing and the meeting kept on the captions. So if you list a client's languages, list all of them — or leave the list empty and let each meeting be worked out on its own.

What do I see if the transcript is still wrong?

A warning, not an invented summary. When the transcript is not coherent enough to summarise, the Summary tab opens with "This transcript could not be read reliably.", gives the reason, and tells you to treat what follows as unverified and check the recording. The notetaker does not fill the gap with a plausible-sounding meeting.

Two other signs are worth knowing:

• When a meeting was translated, the transcript page carries one line: "Translated to English from a conversation held in French", or whichever languages were heard. If that line names a language nobody spoke, the detection went wrong.
• A line the notetaker cannot attribute to a speaker is labelled "Unknown", never guessed.

Can I fix a meeting after the fact?

Partly. Switching a meeting between the Project meeting and Interview profiles, or choosing a different notes language for it, writes new notes from the same transcript — so it fixes a summary, not a transcript.

There is no button to transcribe a meeting again. If the transcript itself is wrong, the recording is still there to play back inside the app, and that is the source to trust. Correction rules, such as a client name the transcript keeps mishearing, apply to future meetings. If your meetings often mix languages, read why meeting captions pick one language: it explains which meetings are most likely to need the audio pass, and why.