For 4Dictate 0.3.225 (beta), last updated 31 October 2026. Android 13 or later; works on phones, tablets and Samsung DeX.
4Dictate is dictation, not voice typing: you speak, it records, and when you stop it turns your speech into finished, punctuated text and types it at the cursor in whichever app you are using, without replacing your keyboard. You can also speak commands: open apps, press Send, scroll, control music, switch 4Zones desktops, switch on a light through 4Home. 4Dictate works on the DeX screen as well as on the phone screen.
The main screen has three tabs (Dictate · Transcriptions · Voice commands) and ⚙ at the top for Settings, whose pages are Speech · Tidy-up · Profiles · Rules & names · Microphone · Keys & buttons · Recording · Grammar · Backup · Connected apps · About.
For a PIN or a code, say "numerical" first: "numerical nine nine nine nine" types 9999, with no spaces and no full stop. Say "point" or "dot" between digits for a decimal or a dotted number: "numerical one point five" types 1.5, and "numerical two dot one dot three" types 2.1.3.
End a dictation with "press send" (or "press enter", "send it") to type it and send it. In a terminal such as Termux, "press enter" presses the Enter key.
To type one dictation in a case, start it with "4Dictate lower case", "4Dictate sentence case", "4Dictate title case" or "4Dictate upper case". The rest of that dictation is typed in that case, after any tidy-up: "4Dictate, upper case, see you on Friday" types SEE YOU ON FRIDAY. You can say "mode" instead of "case" when words follow it ("4Dictate upper mode, see you on Friday"). It changes that one dictation only; every other dictation is typed as normal.
For a terminal, start with "4Dictate command case" (or "4Dictate command mode", followed by the command): everything is lower case, there are no spaces unless you say "space" (or "spacebar"), spelled letters join up, "capital L" types a capital, digit words become digits, and symbols are typed by name ("dash", "slash", "dot", "pipe", "underscore", "tilde", "equals", "open bracket" and more). Nothing is tidied up. "4Dictate command case, adb spacebar devices spacebar dash capital L" types adb devices -L, and so does "4Dictate, command mode, adb, spacebar, devices, spacebar, dash, capital L". End with "press enter" to run it. Said on its own, "4Dictate command mode" still switches to command mode.
When no online tidy-up works on a dictation (tidy-up off, or tidy-up on this device), 4Dictate's basic tidy also lays it out: "new paragraph" starts a paragraph anywhere; "new line" and "bullet point", said on their own, start a new line or a line with •; "number one" up to "number twenty", said on their own, start a numbered line; and an emoji's name followed by "emoji" types it ("thumbs up emoji" types 👍). A reply of three words or fewer gets no full stop, and a search or address box gets no capital and no full stop. An online tidy-up gets your words as you said them and lays them out by its instructions.
If 4Dictate cannot place the text, it says so: a message starting "Not typed" names the app, for example "WhatsApp did not take the text", and the words stay on their card in Transcriptions, tidied, ready to copy. Every recording is kept until you delete it (or for 7 or 30 days: ⚙ → Recording → Keep recordings).
"Scratch that" at the end of a dictation types nothing but keeps its card in Transcriptions. Cancel stops a recording and deletes it.
| Trigger | How |
|---|---|
| Floating buttons | 🎙 dictates, 🤖 takes a command. Tap to start, tap to stop. Drag to move: a button stays exactly where you drop it, kept on the screen and off the keyboard (unless Place buttons anywhere is on, below), and moved just clear if it lands on the other button. Long-press to hide. A small ✈ shows when an offline profile is active. |
| Edit pad | 🎙 Dictate and 🤖 Command keys on the pad (section 4). |
| Keyboard keys | ⚙ → Keys & buttons: a Dictation key, a Command key, a Cancel key, a Paste last key, a "Dictate with" key for each profile, and a 4Home key. Works with Bluetooth keyboards, remotes and volume keys. |
| Gestures | One-handed gesture apps can run 4Dictate shortcuts: Dictate, Command, Cancel, Type last dictation, "Dictate with: <profile>" for each profile, and "4Home: <shortcut>" for each 4Home shortcut. |
Commands can stop by themselves about a second after you stop talking (⚙ → Recording, command pause slider).
Floating buttons (⚙ → Keys & buttons → Floating buttons): switch them on or off, and choose what shows while the device is unlocked: Two buttons, the Edit pad or None. Button size sets both round buttons, from 40 to 96 dp (64 dp to start with). Place buttons anywhere (off to start with): on, a button stays exactly where you drop it, even on the taskbar or over the keyboard, and does not move up when the keyboard (or just its toolbar, with an external keyboard) appears. Leave it off to keep the buttons clear of both. Either way, the two buttons never sit on top of each other. For the lock screen, see section 13.
A slide-out panel for editing what you dictated without the on-screen keyboard.
In a terminal the keys send real key presses (arrows, Home/End, Enter); editing that needs to read the text is not available there. In Termux, Paste (the key, or "paste" in command mode) uses Termux's own paste, so what you copied is pasted.
Use the 🤖 Command button or key, then say the command. The Voice commands tab lists every phrase 4Dictate listens for, lets you add your own (Your commands), shows what was Last heard, and can Share what I can say.
| Group | Examples |
|---|---|
| Send | press send, press enter, send it |
| Navigation | go home, go back, recent apps, notifications, quick settings, take a screenshot, lock the screen |
| Apps | open WhatsApp, open Files (any installed app by name), and your own shortcuts |
| Media | pause, play, next, next track, previous, stop playing, what's playing |
| Scrolling | scroll up, scroll down |
| 4Zones | desktop one … four, next desktop, create desktop, snap to one … four, use layout <name> |
| Modes | command mode, dictation mode |
| Editing (command mode only) | copy that, cut that, paste; select word, select sentence, select paragraph, select all; rewrite that, check that; lower case, sentence case, title case, upper case: the same as the edit pad's keys, on the text box you were in |
| Type it again (command mode only) | paste last dictation, type that again: types your last dictation again at the cursor, as it was typed the first time. The clipboard is not used. Also a key and a gesture (section 12). |
| Profiles (command mode only) | profile <name>, for example "profile offline" |
| Home (command mode only) | each of your 4Home shortcuts by its name, for example "hall light on" (below) |
Tip: choose phrases of two or three words. Single letters or very short words are hard for any recogniser.
You can also give a command while dictating: start with "4Dictate,", for example "4Dictate, press send". Only the start counts; a sentence that merely mentions 4Dictate is typed.
4Home is our companion app for Home Assistant, a paid app. In 4Home you choose which lights, switches, covers, scenes and buttons it may use (only the ones you tick), and each one offers its own Home Assistant actions: on, off and toggle; brightness, colour and white where the light supports them; open and close; press. Tick Shortcut on the ones you want, and give each a name in 4Home, such as "Hall light on" or "Desk up".
The Voice commands tab, under Command mode, then has a Home group listing every 4Home shortcut. Each shortcut's name is its first phrase. Tap a shortcut to add phrases of your own ("hallway on") or to remove one, the name included. If you rename a shortcut in 4Home, the new name is its phrase from then on; your own phrases stay. Refresh reads 4Home's list again.
When nothing above matches, 4Dictate can pass what you said to another app on your device that offers to help, such as 4Tasks: "remind me tomorrow at three to call Sandra", "what's due tomorrow?", "mark the Sandra task done". ⚙ → Connected apps lists the apps that can be asked. Our own apps are used automatically; any other app only after you approve it. 4Dictate asks the app directly, never by tapping the screen. To choose the app and what to ask it, 4Dictate sends what you said, with the apps' lists of what they offer, to your tidy-up service (or the model on this device), even if the profile's tidy-up is off.
Short or long? The 🤖 Command button stops when you pause, which keeps commands quick. For a longer request, or if you think as you talk, use dictation mode: start with "4Dictate,", say everything you want, and press stop when you are done. Nothing from it is typed.
A profile is a set of choices: which service transcribes, whether and how text is tidied up, its instructions, and how strict the Check key is. Examples: "General", "Business" (formal, no slang), "Raw (no tidy-up)", "Offline".
A profile does not set a case: every dictation starts in the normal case. To type one dictation in another case, or as a terminal command, say it at the start ("4Dictate upper case, …"; see section 2).
⚙ → Speech, in the order they are tried:
The models you can use on the device:
| Model | Languages | Download | Best for |
|---|---|---|---|
| Parakeet | English and 24 European languages | about 490 MB | the recommended choice; the most languages |
| Phonon-2 | English only | about 160 MB | fast English: about three times faster than Parakeet on a recent phone, at a third of the size. Needs a 64-bit ARM processor, as almost every recent phone and tablet has |
| Whisper | many languages | your choice (no link in 4Dictate) | languages Parakeet does not have |
| Moonshine base | English only | about 250 MB | older devices; no paragraphs from your pauses |
When you press to start, 4Dictate checks every service at once and gives them one second to answer; one that does not answer is skipped straight away, so a weak connection never leaves you waiting. On slow mobile connections you can allow 2, 3 or 5 seconds instead (⚙ → Speech → Wait for services to answer). Every request also has an overall time limit.
Optional, per profile: an AI model edits each dictation with the profile's instructions: punctuation, capitals, fillers, your style. Your words are kept: a reply that answers you instead of editing, invents text, or drops your sentences is refused and your words are typed as spoken.
The ten minutes you spend on a profile's instructions matter more than any other setting. The built-in instructions are deliberately timid: keep every word, fix spelling and capitals, remove "um". They never sound wrong, but they never sound like you either. Your own instructions are where 4Dictate learns your job, your people and your tone.
Where: ⚙ → Profiles and their instructions → a profile → Instructions. The built-in text is shown in full; change any of it and the profile gets its own copy. Reset to the built-in instructions takes you back.
The same dictation, two sets of instructions:
| You said | "um so tell dave the the quote for the kitchen is four thousand two hundred and we can start on the twelfth" |
|---|---|
| Built-in | So tell Dave the quote for the kitchen is four thousand two hundred and we can start on the twelfth. |
| Your own | Dave, the kitchen quote is £4,200 and we can start on the 12th. |
A starter to copy and adapt:
Edit my dictation into a message I can send. Output only the message.
I run a small building firm in the UK; write UK English.
My clients and crew: Dave Marsh, Priya, Kowalski Plumbing. Spell them exactly.
Write money as £4,200, dates as 12 March, times as 2:30 pm.
Keep it friendly and short. Do not add greetings or sign-offs I did not say.Three tips
4Dictate protects your words whatever the instructions say: a reply that drops one of your sentences or adds text you did not say is refused, and your words go in as spoken. If your instructions are meant to remove asides, switch on Allow tidy-up to remove sentences for that profile; for spoken instructions ("Oracle, add a heading"), set an Instruction word (section 6). Make one profile per kind of writing ("Clients", "Family", "Notes") and switch with a swipe.
Models take time to load (Parakeet ~3 s, Phonon-2 ~1 s, Qwen3 4B ~30 s). Load in aeroplane mode loads them when aeroplane mode goes on; Keep ready keeps them for two minutes after use; an offline profile keeps them loaded while it is active. On a poor connection, switching to the Offline profile is quicker than waiting for online services.
With a speech model on the device, ⚙ → Speech → On this device also offers:
Every recording has a card: what was heard, what was typed, where it went, which profile and services did the work. Search, filter (All, Succeeded, Too quiet, Failed, Lock screen recordings), play the audio, copy, make a rule, delete (with confirmation).
4Dictate does nothing while your device is locked. The floating buttons are hidden, and a gesture or the watch does nothing except show "Unlock to use 4Dictate" on the device. Nothing is recorded, nothing is typed, no command runs and no app is opened. A key you bound to 4Dictate, such as a volume key, does its normal job on the lock screen. When you unlock, the buttons come back by themselves.
The few seconds after the screen goes off, before the device locks, count as unlocked.
If the device locks while you are recording, the recording stops (unless Work while locked is on, below). It is never typed. It is kept in Transcriptions with nothing typed, and you can send it later with Transcribe again once the device is unlocked.
⚙ → Keys & buttons → Floating buttons → Work while locked (off to start with). On, you can dictate while the device is locked: what you say is kept in Transcriptions as a memo. It is never typed, and nothing in it is obeyed: "press send" stays as words. Commands never work while the device is locked, with this on or off.
With Work while locked off, a recording stops when the device locks and is kept untranscribed: use Transcribe again once the device is unlocked.
On a Wear OS watch paired with your phone, 4Watch starts and stops 4Dictate on the phone without your looking at either. The phone's microphone or headset listens, and the words go where the phone's cursor is. Install it from 4Dictate's Play page on the watch.
⚙ → Backup: export and import your settings, profiles, rules and commands, and automatic backups to a folder you choose. API keys are never included in exports: re-enter them on a new device.
| Problem | Try |
|---|---|
| Nothing types, or 4Dictate says the service is stuck | Android Settings → Accessibility → 4Dictate dictation: switch it off and on again. |
| "Not typed", "… did not take the text" | The words are on their card. Another dictation app's accessibility service can take over the typing connection; use one at a time. |
| "Nothing heard" / too quiet | Raise Sensitivity, or speak after the beep with a headset. |
| Slow | Open ⏱ on the card to see which step took the time. |
| Commands misheard | Use two- or three-word phrases; add your own in Voice commands. For terminal words, add them to Command words, or make a Command mode only rule (section 15). |
| "Unlock to use 4Dictate" | The device is locked: see section 13. |
To report a problem, note the time and send ⚙ → Backup → Send diagnostics (no transcripts, keys or rules are included) to [email protected].
There is no account, no analytics and no advertising. Your recordings, and the text sent for tidy-up, Check, Rewrite and connected-app requests, go only to the services you choose, with your own keys, or stay on the device offline. 4Watch sends only "start", "stop" and "send" from your watch to your phone; it records nothing and sends nothing to us. See the 4Dictate privacy policy.
© mr-biz.uk · mr-biz.uk