the best all-round model: fast and accurate, with punctuation and capitals, in 25 european languages.
speed accuracy⌘4part of dipswitch, a mac app
VOICE
you talk faster than you type.
VOICE is dictation, in an app on your mac. hold right ⌥ on your keyboard, say it, let go: the words are typed where your cursor was. it runs on this mac, and every speech model is free.
download free
apple silicon · macOS 13 or later
an ordinary reply
a message that needs more than two words.
a reply in a chat, a paragraph of an email, a note: typing is the slow part. hold right ⌥, say it, let go. it is typed where the cursor is, tidied, in the app you were already in.
- right ⌥hold, talk, let go
- right ⌥ ×2hands-free. tap to finish
- esccancel
- ⌥⇧Vbring a cancelled take back, or type the last one again
the price
every speech model is free. no limit.
no minutes to run out of, no better model kept back, no account. the model that listens runs on your mac, not on a server, so there is nothing to meter.
speech models · all free
for everything else: about 99 languages, and better than STANDARD in polish, dutch, swedish, finnish and danish.
speed accuracya small, quick english model: the fast start while STANDARD downloads.
speed accuracy- first run
- one download before your first dictation
- 11 s of speech
- 54 ms on an M4 Max the standard model’s published figure. the app measures yours.
the tidy-up
um. gone.
capitals and punctuation go in. um, uh, er and hmm come out. tell it a name once and it writes it your way. clean-up and the word list are free too.
as saidum running ten minutes late uh start without me
typedRunning ten minutes late, start without me.
- word list
- write it as Siobhán, when it sounds like “shiv on”
- say
- “new line”, “new paragraph” english only for now.
- lands
- typed at the cursor, or copied for ⌘V if you prefer
what is kept
it does not keep your voice.
the microphone is open while you hold the key, and only then. your speech becomes text on this mac and the audio is never stored. no cloud, no apple speech.
- recent
- your last 20 takes, as text, for up to 24 hours. gone when you quit the top rows are the takes from this page.
- password fields
- not typed, and never shown. ⌥⇧V types it anyway if you meant it
- downloads
- the model files, once, from hugging face
said plainly
what it needs, and where it stops.
- microphone
- macOS asks once
- accessibility
- needed to type into other apps and to hear a key on its own without it your words are copied for ⌘V, and ⌥ space is the key.
- secure input
- while a password field has the keyboard, VOICE cannot hear your key
- language
- detected for each take it can mistake polish for russian. pin one language and it stops.
- airpods
- skipped unless you pick them, so your music keeps its quality
- other keys
- either ⌥, left ⌥, right ⌘, right ⌃, fn or ⌥ space
- pause
- 15 minutes, an hour, or until you resume. the key is a normal key again
- PRO
- edit by voice, a style per app, learning from your fixes bought once, no account. dictation stays free.
download free.
what free does
- DISK
- scan, map and rate the whole drive. build the delete list.
- NET
- every app and where it connects, live and for today.
- PASTE
- history, search and every privacy rule. seven days kept.
- VOICE
- dictation with every speech model. no limit.
- LOAD
- every view, today’s history, and quitting what is in the way.