Private audio transcription

Speak. Etch. Done.

Offline voice notes that transcribe on your Android phone

Someone walks you through the whole plan over lunch. You nod along, and later you realise you only half remembered it. With Etch you just hit record. When you are done you have the full words, a short summary, and a clear list of what you need to do.

$3.99, once. No subscription, no account, no ads.

Transcribes on your phone
No account
No subscription
Works offline

Mock appointment, real transcript. 1 minute 33 seconds.

For the moment someone talks faster than you can write

Some conversations only happen once. Somebody catches you on your way out, or you get five minutes with a person who is already late, and you’re told four things you need to remember in under a minute.

A colleague with the whole plan, over lunch
A boss with four tasks in a hallway
Someone showing you how a thing works, once
Car shop or home repair
The bank or the insurance office
Parent–teacher conference
You, in the car, saying it out loud before you forget it
Doctor, dentist or pharmacy, for your own notes

Install it once so it is there the day you need it.

“The doctor spoke. I nodded. I had nothing.” 39 seconds.

The conversation you meant to write down. 38 seconds.

1
Talk
Tap record and say the thing before you lose it. It keeps recording when the screen goes dark, so you can put the phone down.
2
The phone does the listening
Your phone turns what you said into text right there on the device. You do not need internet, and you do not need an account.
3
Get something you can use
When it is finished you have a transcript you can search and correct. If you want them, you also get a short summary and a checklist (Action items, in the app).

Interactive demo · not a screenshot

Try it here first

A grocery list is already on the phone. Tap Record and a 13-second stand-up update plays (sound on). Then do what you would do in the app: open it, play it back, fix what the speech model got wrong, tag it, and ask for notes, a summary and a checklist.

Find the comma that shouldn’t be there, and fix it with Edit
Generate the notes before and after the fix
Rename Priya or Marcus and watch the notes follow
Tag the stand-up, then filter the list by tag

Both transcripts are exactly what Etch wrote from these clips, mistake included. The notes and summaries came back from a small AI model we ran them through, and so did the grocery checklist; the stand-up checklist we wrote ourselves, in the same format. Nothing here leaves this page.

Watch the real setup ↓

Try it right here

Watch someone set it up

Etch, Getting Started. 2 minutes 55 seconds.

Create your free OpenRouter account →

What you will see, in order:

The safety notes you agree to
The speech model downloading, which happens once and not every time
A recording turning into a transcript

All of that works straight away, with no account and nothing to set up.

Then setting up your own AI (the fiddly part), which you only do once:

Getting a free OpenRouter key in a browser and pasting it in
Picking a model and testing that it works
The same recording turning into notes, a summary and a checklist

That setup is not repeated. Every recording after it goes straight to the summary.

Your voice stays on your phone

The phone does the transcribing itself, using a speech model Etch downloads once. Your recording is never sent anywhere.

That means it works in a basement, on a plane, and on a train going through a tunnel. It also means nobody at NeoVetta can read or hear your recordings, because there is no NeoVetta server for them to sit on.

Etch home screen with a record button and a list of transcribed notes

Summaries and tasks, not a wall of text

A ten minute ramble becomes three lines you will actually read again, plus a checklist of the things you said you would do.

This part is optional, and it uses an AI service that you choose. Only the text of the transcript is ever sent. The audio never is, and nothing goes anywhere except the service you set up yourself.

Notes, summary and action items generated from a stand-up recording

Correct a name once and it stays corrected

Every speech model trips over unusual names. Etch heard NeoVetta as “NeoVetto”. So the transcript is editable, and find and replace is built in. Fix it in one place and every other mention changes with it.

Ask for the summary again afterwards and the correction carries through to your notes and your checklist. Most apps either make you live with the mistake or run the same recording again and hand you the same mistake back.

The transcript editor with Find, Replace with, and Match case fields

Your AI. Your key. Your bill.

If you want summaries, you decide which AI service writes them, and you pay that service directly. Etch does not sit in the middle and does not take a cut. Most of the well known services work, and so do a lot of smaller ones. You can even point it at a model running on your own computer, in which case nothing leaves your house.

There is a free way to do this. An OpenRouter account costs nothing and its free models do not ask for a card. Etch says plainly in the app that free services usually train on whatever you send them, because that is worth knowing before you send anything.

Provider settings showing OpenRouter, an API key field and a live model list

Offline transcription, compared

The best-known transcription apps, such as Otter, Notta and Transkriptor, upload your audio and transcribe it on their servers. Etch transcribes on the phone. Here is what that changes.

	Etch	Cloud transcription apps
Where the audio is transcribed	On your phone	On their servers, after an upload
With no signal	Records and transcribes	Records, transcribes once back online
Account	None	Required
Price	$3.99, once	Usually a free tier with limits, then a subscription
Summaries	Optional. Transcript text only, sent to the AI you choose	Built in, on their servers

When a cloud app is the better choice

You need a language other than English. Etch is English only for now.
You want the words on screen while people are still talking. Etch transcribes after you stop.
You need the transcript to say who said what. Etch does not label speakers.
You want it on a laptop too, or a bot that joins your video calls. Etch lives on one Android phone.

If none of those matter and you would rather your recordings never left your hands, Etch is built for exactly that.

The rest of it

Play it back

Jump to any moment, skip forward ten seconds, or slow it down. Useful when you want to hear exactly what was said.

Import audio

Already have a voice memo or an interview recording? Bring it in and Etch transcribes it the same way.

Export

Share or save the original recording, or a smaller version if you are sending it to someone. It is your recording and your file.

Long recordings

Keeps recording with the screen off, and breaks a long session into separate files so no one file gets unwieldy.

Light and dark

Cream and forest, or forest and cream. Pick the one you prefer and it stays that way.

Straight answers

Does my audio ever leave the phone?: No. The transcribing happens on your phone every time, so there is no upload step. There is also no NeoVetta server to upload it to.
What about the AI summaries?: Only if you set them up. Then the text of the transcript goes to the service you chose. The audio never does. What that provider does with it is governed by their policy, not ours.
How do I set it up?: Recording and transcribing need no setup at all. If you want summaries you add an AI provider once, which is the fiddliest part. There is a three minute walkthrough that shows the whole thing.
Do I need an account?: No. There is no login, no profile and no analytics of any kind.
What is the local transcription model?: A Whisper speech model, downloaded once the first time you use Etch and then kept on the phone. It is the full model rather than a cut-down version that sends the hard parts off to a server, which is why transcription keeps working in a basement, on a plane, or with the phone in airplane mode. English only for now.
Will this work with my AI provider?: Almost certainly. OpenRouter, OpenAI, Groq, xAI, DeepSeek and Mistral all work, and so does anything else that speaks the same OpenAI chat format, which most services now do. You paste in a key, pick a model, and Etch tests it for you. It also works with a model running on your own computer, in which case nothing leaves your house at all. Summaries are the only part that uses a service like this. Recording and transcribing never do.
What languages?: English only for now. The speech model on the phone understands English, and if you give it another language it will confidently invent something rather than tell you it cannot cope. More languages are the next thing we want to build.
How is this different from Google Recorder?: Google Recorder also does its transcribing on the phone, and it is very good, but it is only available on Pixel devices. Etch does the same job on any 64-bit Android phone from 8.0 upwards, and adds the summary and the checklist.
Can it record phone calls?: Not a call happening on the same phone, because Android does not let any app do that. Etch records what the microphone hears, so a call on speakerphone works, and so does anything being said in the room.
Which phones?: Any 64-bit ARM Android phone running version 8.0 or newer, which is nearly every Android phone sold since about 2017. It will not run on an emulator.
How much?: $3.99, once. No subscription, no in-app purchases, no ads, and no upsell later. If you point it at a paid AI provider for summaries, that bill is between you and them. We never see it.
How long does transcribing take?: Roughly as long as the recording itself, and faster on a newer phone. It keeps going with the screen off, and shows you how long it has been working.

Get Etch

Etch is on Google Play for $3.99, once. There is no subscription, no account, no ads and nothing to cancel later. It works in English only for now, and runs on 64-bit ARM Android 8.0 or newer.