Minutes of a meeting, or the sentence you are halfway through
Otter.ai records meetings and long conversations, uploads the audio, transcribes it in the cloud and hands you a searchable transcript afterwards, tied to an account. That is a real job and it is the job Otter is built for.
LocalType is not that. It is an Android keyboard with a microphone key. Speech becomes text as you write, straight into the field you are writing in, and the app does not record meetings, produce transcripts of conversations, or keep anything once the words have landed.
So if what you need is "somebody took the minutes for me", no keyboard replaces that, and you can stop reading with our blessing. LocalType is a separate product and is not affiliated with Otter.ai.
This is for the other reader: the one who searched for an alternative because of where the audio goes, and who was never really after a transcript in the first place.
Booking a meeting room to answer a question
A surprising amount of what people bring to a transcription service is not a meeting at all. It is:
- dictating a long message you do not fancy typing
- getting an idea down before it evaporates, into whatever note app is closest
- writing an email on a phone without hating the process
- filling in the same form fields you fill in every week
- replying properly instead of sending three words because typing is tedious
Each of those happens inside an app you were already in, needs the text where your cursor is, and is finished before a transcription tool would have opened. Strip away the meeting features and Otter's core is speech becoming text, which is the same core a keyboard has. The difference is the route the audio takes to get there: out of the device, onto servers, and back as a result that lives in an account.
For everything except recorded meetings, that route is avoidable. Speech recognition is software, and it runs perfectly well on the phone in your hand instead of on hardware you will never see. LocalType does exactly that, wherever a keyboard opens, on phones and tablets running Android 8.0 or newer.
Picture the practical shape of it, because it is unremarkable. You are in a messaging app, halfway through a reply, and the next bit is a sentence you would rather say than thumb out. You press the microphone key on the keyboard that is already open, say it, and carry on typing the word after it. Nothing was opened, nothing was copied, the clipboard still holds whatever it held before, and there is no second app waiting in the background holding your microphone.
Three things a keyboard will never give you
A transcription service does three things a keyboard will never do, and no list of advantages changes any of them.
A document. Otter leaves you a transcript you can search, revisit and share six months later. LocalType produces no artefact at all: no file, no history, no timeline of who said what. The text went into your message and the process ended. That is the design goal when you are writing and precisely the wrong one when you are documenting.
More than one voice. Otter handles several people in a conversation, some of them turned away from the microphone, some talking over each other. A keyboard transcribes one voice, yours, close to the phone, aimed at it deliberately. The models that fit on a handset are not the ones you would pick for a table of four people.
Length. A keyboard works in bursts. You speak a sentence or a paragraph, the words land, you carry on typing. Nothing about it is built to sit in the corner of a room for forty minutes, and using it that way would be a choice you regret rather than a limitation you discovered.
If your problem is minutes, notes from calls, or interviews you need to search next year, keep using something built for it. The trade on offer here only makes sense for the writing you do in between meetings.
There is a workable arrangement in the middle, and people arrive at it without being told to. The recorder handles the calls where somebody has to be able to prove what was agreed. The keyboard handles the follow-up message you send twenty minutes later, which is the part that was never going to be minuted and was always going to be typed with one thumb on a train.
Storage, memory, and eighteen languages out of one file
The model has to be on the device, so it occupies storage and holds memory while it works. That is the real cost of not sending audio anywhere, and it does not go away.
Three sizes ship with the app. Compact is 60 MB and returns text quickest, which suits an older phone. Standard is 190 MB, is what the app recommends, and is the right answer for most handsets. Large is 539 MB and reads accents and background noise best, at the cost of waiting: on a test phone with 4 GB of memory, Standard converted 6.9 seconds of speech into text in 4.1 seconds, and Large takes roughly twice as long as you spent speaking. Only one is installed at a time, so trying another simply replaces it and nothing piles up behind you.
Those eighteen languages come from one multilingual model, and LocalType works out which of them you are speaking from each recording, so nothing has to be declared in advance. Switch language between messages and it keeps up. Fix it to a single language and the keyboard layout, the dictionary and the dictation language move as one. Accuracy varies across the eighteen, because training material does, and the widely spoken languages come out best. That is true of every speech model, including the ones running in a data centre.
Fetching the model is also what the app asks for internet access for, and it asks for exactly one purpose. Afterwards dictation works with the connection off, on a plane or in a basement, because there is no second use for the network anywhere in the product.
No account exists. No sign-up, no login, no profile, nothing connecting the keyboard to a person. The product carries no advertising and no behavioural tracking runs inside it. No archive of your recordings is kept, since the audio becomes text and then stops existing. The model is a file in storage no other app can open, the app's data is deliberately kept out of Android's cloud backup so it is not quietly copied off the device that way, and in password fields the microphone key is not offered at all.
The rest of the app is an ordinary keyboard, and that is deliberate. Autocorrect and next-word suggestions come from a dictionary matched to the language you set, the standard layouts are there including QWERTZ and AZERTY, and you can add your own surnames and jargon by hand. Those additions stay on the handset and are handed to the speech model as context, so the names you actually use start being spelled correctly.
Meetings, or the rest of the day
Use a transcription service for meetings and recordings, where the value is in the document it leaves behind and where an account is the price of having one.
Use LocalType for the writing you do all day, where you want the words in the message and nowhere else, and where the answer to "what happened to my voice" should simply be that it stayed on the phone.
Plenty of people end up running both, and there is nothing contradictory about it. One handles the conversations you have to document. The other handles the hundred small things you would rather say than type, without any of them leaving the handset.