Developer of GigaAM-Powered Offline Dictation App Brings Pisar to Google Play, Asks for Testers

Giga Pisar is an Android dictation app that turns speech into text directly on the device. It works on a push-to-talk basis: you hold a button, speak, and release, and the text appears wherever the cursor is. Recognition is handled by Sber's GigaAM model, with no internet and no cloud. Until now, the only way to install the app was an APK file from the author's website. He has now decided to publish Pisar on Google Play, and immediately warned that there is a catch: the release can't happen without help from readers.

What Pisar can do

The idea is simple, which is why it works. The app doesn't make you switch between windows, copy the result, or open a separate editor. It inserts the recognized text into the active field: in a messenger, a notes app, or a browser. To do this, it needs access to Android's Accessibility features.

Its main selling point is local processing. Audio never goes to a server, so dictation works on the subway without a connection and doesn't send personal voice messages to third parties. For Russian speech, this matters: GigaAM was developed for the Russian language, and running it on a smartphone turned out to be realistic.

Why installing from a file is a pain

The author is candid about the problem. Installing the APK from the website works, but it looks suspicious. The phone warns about an "unknown app." To enable Accessibility features, you have to go into "restricted settings" and manually lift the block. For a technically savvy user, that takes a couple of minutes. For everyone else, it's a reason to close the page.

For an app that already needs a sensitive level of access, this is a double problem: the audience's distrust compounds the inconvenience. An app store removes some of those barriers: one-tap installation, automatic updates, and a familiar vetting channel.

What's the catch with Google Play?

In the publicly available part of the announcement, the author doesn't say exactly what the difficulty is. All we know is that he needs help from readers. Based on the experience of other indie developers, Google's rules for new personal accounts usually require closed testing: at least 12 testers must use the app for 14 consecutive days before publication opens up. That would explain the request for help, but the available excerpt doesn't confirm it, so this is only a guess.

An additional complication for apps of this type is the declaration for using the Accessibility API. Google requires developers to explain why a service needs Accessibility features, and it checks this more strictly than ordinary permissions.

Key project parameters:

• Recognition model: Sber's GigaAM
• Mode of operation: fully on-device, no internet and no cloud
• Controls: hold the button, speak, release
• Output: text is inserted at the cursor position
• Current distribution: APK from the author's website
• Goal: publication on Google Play, free of charge

How to help and what to watch for

If you use Android and are interested in voice input, it's worth reading the original article on Habr, where the author explains exactly what kind of help he needs. Taking part in closed testing usually requires a Google account and a couple of minutes to sign up.

Users who decide to install the APK right now should keep in mind that an app with Accessibility access can see what's on your screen. Check the source and verify the file's checksum if the author publishes one. With the Google Play version, this risk is partly reduced by external review.

This story is telling for the market. Local speech recognition models are already compact enough to run on a phone. For an independent developer, the hardest stage turns out to be not the neural network but getting through the app store.