Change transcription accuracy or language

In Settings, open Transcription.

Language offers English and Multilingual. The multilingual model covers 25 languages, English included. Meetings and dictation each have their own language setting.

With English selected, the Accuracy slider has three stops:

• Fast (a 45 MB model): lowest CPU use, for older machines.
• Balanced (73 MB): the default.
• Accurate (662 MB): highest accuracy, higher CPU use. This is also the only local option that labels speakers live during a meeting.

A marker shows the recommended stop for your machine’s memory. Picking a stop whose model isn’t on disk prompts the download. The Models settings section manages every downloaded model in one place.

The cloud transcriber ignores these choices and uses its own models. See Local and cloud processing.