Skip to content

Choosing a speech model

Four engines, all on device. Bigger is more accurate and slower; the default is the one most people should stay on.

Settings, Transcription lists the engines. Picking one downloads it once, to your Mac, and dictation then runs with the wifi off.

The four engines

Fast is the smallest and the quickest to start. Accurate is the default and the right answer for most people: it makes far fewer mistakes on names. Max Accuracy is the largest Whisper build and worth it only if you dictate difficult audio. Parakeet v3 is the fastest of the group for English and 25 European languages.

Compact           574 MB
Accurate          874 MB   (default)
Max Accuracy      1.6 GB
Parakeet v3       670 MB

Where the models are stored

Models live in Blurt's folder inside Application Support. Deleting one frees the space; Blurt re-downloads it if you pick that engine again.

~/Library/Application Support/ → the Blurt folder → models
Switching engines does not change your vocabulary, replacements or history. Those are shared across all four.