The transcript is in the wrong language

Canto detects the spoken language automatically from the audio itself, and there is currently no way to set the transcription language manually. When detection gets it wrong (usually on very short recordings or mixed-language speech) retrying the transcription is the one lever you have.

Why does auto-detection miss?

Detection works from the speech in the recording, and it's very reliable when there's enough of it. It struggles when there isn't much to go on:

• Very short recordings: a few seconds of speech may not be enough to identify the language confidently, especially between closely related languages.
• Mixed-language conversations: meetings that switch between languages, or use heavy loanwords from another language, can tip detection toward the wrong one. The transcript follows a single detected language, so the minority language's passages come out garbled.
• Quiet or noisy audio: when the model can barely make out the words, its language guess suffers too. See Get better audio quality.

Canto transcribes the languages listed in Which languages does Canto support?; a language outside that list can't be detected correctly no matter how clear the audio is.

Can I set the language manually?

No. There is no manual language override for transcription; detection is always automatic. We're honest about this because retrying in the hope of finding a hidden setting wastes your time; if your recordings are consistently detected wrong, tell support which language pair is involved so we know where detection is failing.

Note that your app language setting is not that override: it controls the interface and the language of AI-written content like summaries, but has no effect on what language the transcript is in. See Change the app language.

What can I do about a wrong-language transcript?

1. Open the recording and choose Retry transcription from the ⋯ menu. Detection runs fresh each time, and borderline recordings often come out right on a second pass. See Retry a failed transcription.
2. If the same recording keeps detecting wrong, the audio itself is the problem: too short, too mixed, or too unclear for a confident call. The audio is unchanged and safe, so a retry never makes things worse.

How do I avoid it next time?

• For short notes, speak a full sentence or two rather than isolated words; detection needs material.
• If a meeting will run mostly in one language, let the recording open with that language rather than greetings in another; the opening minutes carry weight.
• Record close to the speakers so the speech is clean.

The transcript language is right but the summary isn't

Summaries and other AI-written content follow your app language, not the transcript's language. If the summary comes out in a language you don't want, change the app language setting; see Change the app language.