Two jobs, one search
Live Transcribe is an accessibility app. It listens to the room and puts what is being said on the screen in real time, so that somebody who is deaf or hard of hearing can follow a conversation. It is free, it is built for that purpose, and on suitable devices it can work without a connection.
LocalType does something else entirely. It is a keyboard: you press a microphone key, say a sentence deliberately, and it becomes text in the field you were typing in.
The first is about understanding other people. The second is about producing your own writing. If you need the first, nothing here replaces it, and you should stop reading and install the app built for the job.
No affiliation exists with Google or with the team behind that app. LocalType is a separate product.
Whose voice, and pointed at what
Three questions settle it, and the third one is the one the engineering follows from.
Whose speech is it? Other people's, in a room, unprompted, means captions. Your own, spoken on purpose into a phone, means dictation.
Do you want to read it or send it? Captions exist to be read as they happen. Dictation exists to produce text that goes somewhere.
Is the microphone pointed at you or at the room? Captioning is far-field, multi-speaker and continuous. The microphone is picking up a room, several people may talk at once, nobody is enunciating for a machine, and it has to keep going indefinitely. Dictation is near-field, single-speaker and bounded: one person, close to the phone, speaking deliberately, in short bursts. The second problem is much easier than the first, which is why it fits comfortably on a handset.
Ask a keyboard to caption a meeting and it will do a bad job, because that was never the design. Ask a captioning app to write your emails and you will be copying text out of it all day. Most people arriving here from a search have the second need and reached for the first tool because it was the transcription app they had heard of.
If you are choosing for somebody else
People often arrive at this search on behalf of a parent, a partner or a colleague, which makes the distinction above harder to apply, because you are guessing at somebody else's difficulty.
The question to ask them is not which app they want. Ask instead whether the problem happens when other people talk, or when they themselves try to write. Those are different frustrations and people describe both as "I have trouble with my phone".
If it is the first, captions. If it is the second, a keyboard with dictation. And in either case do the setup rather than handing over a name: install it, enable it in Android's settings, download whatever it needs while you are on Wi-Fi, and type in the twenty names they say most often. An app left half-configured on somebody else's phone gets abandoned on the first evening it does not work.
If you want to produce text
For the reader who wanted dictation rather than captions, the gap is practical rather than technical.
A keyboard's whole value is that there is no transfer step at all. The words appear where your cursor already was, inside whatever app you were using, so a reply takes as long as saying it. With any separate transcription app, even a good one, the loop ends with selecting, copying, switching app and pasting.
Recognition also runs on the handset here, from a model kept in storage only this app can reach. That has the same practical benefit as an offline captioning mode, except that it applies always, not only on supported devices in supported circumstances: no connection is involved at any point after the model download. Three model sizes are offered, at 60, 190 and 539 MB, so you set speed against accuracy yourself, and on the 4 GB test handset the middle option turned 6.9 seconds of speech into text in 4.1 seconds. Eighteen languages come out of one multilingual model, identified from the recording rather than from a menu.
Internet access is requested for exactly one purpose, collecting the speech model you chose, and the app never reaches outward again afterwards. Nothing needs registering. Advertising is absent, and no behavioural tracking runs inside the product. No archive of what you said is kept anywhere, since the sound becomes text and its life ends there. The app's data is held back from Android's cloud backup on purpose. In a password field the microphone key is simply not offered.
Underneath the microphone it is a keyboard doing the ordinary things: autocorrect and next-word suggestions from a dictionary matched to your language, the layouts people expect, a full emoji set, and space to add your own names and jargon. Those additions stay on the phone and are given to the speech model as context, so the words you use most start arriving spelled correctly.
Where captions and dictation genuinely meet
One overlap, and it is a real use.
People who use captioning to follow conversations often also need to reply, and replying is a typing problem; the listening side is already handled. For anyone who finds typing slow or difficult, dictation on the reply side pairs naturally with captions on the listening side, and the two do not compete for the same slot on the phone.
Captions for input and dictation for output is a common arrangement, and running both is the ordinary case rather than an unusual one.
In practice the two sit in different places on the phone. The captioning app is something you open when a conversation starts, full screen, doing one thing. The keyboard is not opened at all; it appears when you tap a text field, in the messaging app you were already using. Nothing has to be switched between them, because only one of the two is ever a destination.
Not a captioning app, and not assistive technology
Four things it does not do.
LocalType does not caption conversations, transcribe recordings, listen continuously, or produce a transcript document. The microphone opens when you press the key and closes when you stop, and between those moments nothing is listening. Multiple speakers are not handled, and it is not built for a microphone sitting in the middle of a table.
It is also not accessibility software in the sense Live Transcribe is. It has not been tested against screen readers or other assistive technology, and this page makes no claim about how it behaves alongside them.
So if you need to follow what is being said around you, use the tool built for that. It is free, it is designed for the purpose, and we are not competing with it. If what you need is to write more easily on a phone, using your own voice, into the apps you already use, that is a keyboard's job. Pick by whose voice it is.