Which speech model should I use?
Base suits almost everyone, Small is noticeably better on accents and long dictation at about a third the speed, and Tiny trades accuracy for size and memory.
Dictate Safe offers three speech models, all of which run on your own computer. Base is the default, and the right choice for most people. Pick Small if you hear frequent mistakes, especially with an accent or a long dictation. Pick Tiny only if you need the smallest download.
#The three, compared
- Tiny (78 MB, 4 GB of memory). The fastest and smallest, and the least accurate of the three. Fine for short, clear dictation; expect to correct more on accents, background noise and long dictation.
- Base (148 MB, 8 GB of memory). The balanced default. A dictation that took you 60 seconds to say is transcribed in about 4 seconds on a 2018 Intel MacBook Pro.
- Small (488 MB, 16 GB of memory). Better on accents and long recordings, and about a third the speed of Base on the same machine. Worth it if Base keeps missing the way you say things.
Timed on a 2018 Intel MacBook Pro, the oldest kind of Mac Dictate Safe supports, so a newer computer is usually faster than this.
#Changing your model
- Open AI Models in the sidebar.
- In the Speech card, press Download next to the model you want, if it is not already there.
- Press Use this. The one in use is marked In use.
Your next dictation uses the model you chose, with nothing to restart.
While Code Mode is on, Dictate Safe uses Small if you have downloaded it, because it is much better at symbol names. Your own choice is what ordinary dictation goes back to.
A model your computer does not have enough memory for is marked Needs more memory on the AI Models page. Switching models does not change anything already in your History; only new dictations use the new one.