Why is AI cleanup slow?
Cleanup runs only on the sentences around something to fix: a few seconds on a 2018 Intel MacBook Pro, up to about thirteen seconds on a long dictation.
AI cleanup only runs when a dictation has something to fix: a spoken correction, or a word close to one in your Vocabulary. Then a language model on your computer reads the sentences around it. On a 2018 Intel MacBook Pro that takes a few seconds, and up to about thirteen on a long dictation. A dictation with nothing to fix costs nothing extra.
#What decides how long it takes
- Whether there is anything to fix. Most dictations have no correction in them and never reach the model.
- How many corrections there are. The model sees only the words around each one, not your whole dictation, so the cost follows the number of corrections rather than how long you talked.
- Your computer. Apple Silicon and newer PCs run it faster than the 2018 Intel machine these figures come from. The Test speed button on the AI Models page times it on yours.
#Making it faster
- Use a cloud provider for cleanup with your own key. It is faster on an older computer, and your voice still stays on your machine: only the finished text is sent. See using your own API key.
- Turn it off in Settings → Cleanup. Filler words, spoken punctuation and your Vocabulary’s spelling are still handled instantly, without it.
Dictate Safe checks whether cleanup can run before letting you switch it on: the part of the app that runs the model, a downloaded cleanup model, and enough memory. If any of those is missing, the switch stays off and the row says which one. See running cleanup on your own machine.
Cleanup is skipped entirely while Code Mode is on, so it never adds time to dictating code.