Two ways into the same job
Phravia is a dictation app for Android and Windows. Its microphone drops tidied text into whatever you were already using, with punctuation, capitals and the ums taken care of, and it does not ask you to create an account. It leaves your keyboard alone.
LocalType comes at the same job from the other end. It is the keyboard, with a microphone key on it, and the speech model that turns your voice into text sits on the phone and never has an alternative.
For the record, LocalType is a separate product with no affiliation to Phravia or its developers.
Windows, and the cleanup pass we do not do
Two reasons to stay exactly where you are, and neither of them is small.
Phravia runs on the desktop as well. It is available for Windows alongside Android, so one dictation habit covers the machine you work on and the phone in your pocket. LocalType is an Android keyboard and nothing else. If your day is split between a laptop and a phone and you want the same tool on both, that is a genuine advantage and no amount of arguing changes it.
Phravia tidies up what you said. Their site describes cleaning the transcript before it lands: punctuation and capitals sorted, filler words removed. That is a real feature and it is not one LocalType claims to have. Our speech model produces a transcription of what you actually said, dropped into the field you were typing in, and any tidying after that is yours to do.
Whether the second one matters depends on how you dictate. If you think in complete sentences and say them once, a cleanup pass is solving a problem you do not have. If you circle a thought three times before landing it, that pass is doing real work on your behalf.
A switch you have to remember the position of
Phravia's design gives you a choice: cloud transcription when you want speed, local transcription for the languages it supports when you would rather keep things close. That is a defensible way to build a dictation app, and for plenty of people the flexibility is the whole feature.
A switch has a cost, though, and it is not a technical one. It is that you now have a thing to remember. Which side is it on right now? Did it stay there after the last update? Is this particular sentence, the one about a colleague or a diagnosis or a password hint, going out over the network or not?
Our answer is to have nothing there to switch. No cloud transcription exists in the app, so the state can never be wrong, and there is no combination of settings that sends a recording anywhere. Whether that is better depends entirely on whether the flexibility was worth anything to you.
Beside your keyboard, or as your keyboard
The second real difference is where the microphone lives.
Phravia sits next to the keyboard you already have and hands text over to it. Nothing about your typing changes.
Ours replaces the keyboard entirely, so the microphone is one of its own keys. You are writing, you hit a sentence that is easier said than thumbed, you press the key, you say it, and the next word you type carries on from where the dictation stopped. There is no second app holding the microphone and no handover between the two.
The upside of Phravia's approach is that you keep the keyboard you know. The upside of ours is that dictation and typing were designed against each other: one language setting drives the layout, the dictionary and the dictation together, and a name you add to the dictionary helps both. Any app where a keyboard opens is covered, on phones and tablets running Android 8.0 or newer.
Where our part ends
One thing is true of both designs, and it is the limit of what either product can promise.
Once the words land in a message, a note or a form, they belong to that app. Our part ends at the moment the text appears in the field you were typing in, exactly as if you had typed it yourself. What the messaging app you send it through does with it afterwards is governed by that app, not by the keyboard that helped you write it.
Sixty, 190 or 539 megabytes of your own storage
Local transcription is not free, and the cost is easy to leave out of a comparison. The model that recognises your speech has to physically be on the phone, where it occupies storage and holds memory while it runs. In a cloud setup that file lives on hardware you will never see, so nobody ever has to think about it.
We make that cost visible instead, and let you choose how much of it to pay. Compact is 60 MB and the quickest to answer, which is what an older handset wants. Standard is 190 MB and suits most phones, so it is what the app proposes. Large is 539 MB and handles accents and noisy rooms best while asking you to wait: on a 4 GB test phone, Standard converted 6.9 seconds of speech in 4.1 seconds, and Large takes roughly double the time you spent talking.
The phone holds one of them at a time, so trying a different size just swaps the file out and leaves nothing stacked on top of it. Downloading it is the single thing the app asks for internet access for, and it asks for exactly one purpose.
What you buy with that storage is a keyboard that does not care about connectivity. Flights, tunnels, foreign SIM-free weeks, thick-walled buildings: none of them are edge cases, because the network was never in the loop after the initial download.
Letters, layouts and the words you add yourself
Ninety per cent of the app is an ordinary keyboard.
Typing is the main job. Autocorrect and next-word suggestions come from a dictionary matched to your language, QWERTY, QWERTZ and AZERTY are all there where those are the national standard, the emoji set is complete, and there is room to add your own names and jargon by hand.
Those added words stay on the handset and double as context for the speech model, so a surname you taught it starts appearing spelled correctly when you dictate as well as when you type it. That is the practical argument for one app doing both halves: the vocabulary you build up serves the keyboard and the recogniser at the same time, and you only build it once.
Dictation, and the nothing it leaves behind
Dictation is a key on that keyboard. The speech model is a file kept in storage only the app can read, and it is the only thing that ever hears you.
There is no account anywhere in the product, so nothing ties the keyboard to a person, and no sign-in step exists to skip. No archive of your recordings is kept, because there is nothing to keep it in and nowhere to put it. The app carries no advertising and no behavioural tracking runs inside it. Its data is deliberately excluded from Android's cloud backup, which has the awkward consequence that a new phone starts blank and you download the model again. Tap into a password field and no microphone key is offered there at all.
None of that is a mode you switch on. The product is deliberately narrow, and the narrowness is what makes that list short enough to check.
Eighteen languages, decided per recording
Local support that covers only some languages is a real limitation of any on-device approach, and the exact list matters more than the word multilingual.
One multilingual model covers eighteen languages, and you download it once. It infers which of them you are speaking from the recording itself, so moving from one language to another between messages needs no menu. Fix it to a single language instead and the keyboard layout follows along.
Accuracy across those eighteen is not uniform. Some languages simply had more training material than others, and the well-resourced ones show it.
One app or two
If you are attached to your keyboard and want dictation added beside it, with cloud speed available when the sentence is harmless, Phravia is built for exactly that and does it deliberately.
LocalType is for the reader who would rather not be asked. One app instead of two, the microphone on the keyboard you already type on, and a single answer to the question of where your voice goes.