How to use LiveCaptionN

Allow two permissions

Open the app. In the Permissions card tap Grant next to Microphone and Overlay (Android calls it “Display over other apps”). The caption window cannot float above other apps without the overlay permission, and the mic cannot be heard without the microphone one.

The Start button stays grey until both are granted and a speech model for your source language is installed. English and Vietnamese models come built in, so on most phones Start lights up as soon as the permissions are given.

The System Compatibility Check card tells you whether your phone can run on-device recognition. “Fully Supported” means you can leave everything on the defaults.

Pick where the sound comes from

Audio Source is the first real choice:

• Microphone – listens to the room: a person talking next to you, a TV, a meeting, a lecture.
• System Audio – captures the sound of apps playing on the phone itself (a video, a podcast, a call-in app) without the microphone. When Android asks, choose Share entire screen. Some apps do not allow their audio to be captured; if the window stays empty, switch back to Microphone and play the sound out loud.

Below it, Transcription engine is where speech is turned into text. Local Vosk is the default and runs entirely on the phone. With Microphone you can also pick Android built-in (the phone’s own recognizer); with System Audio you can pick Remote Whisper (a Whisper server you host, see the remote setup guide). If you are not sure, keep Local Vosk.

Check the speech model (optional)

Tap Manage on-device models. The Installed list shows the Vosk models already on the phone – English and Vietnamese are bundled, about 40 MB each. You only need a model for the language that is being spoken.

Small models are quick to download and light on storage. Large server-grade models (80 MB to 2 GB) are noticeably more accurate and are part of Pro. Downloading is a one-time step; models stay on the phone and work offline, and you can uninstall any of them later to free space.

Set the two languages

In the Languages card:

• Source (spoken) – the language people are speaking. With Local Vosk this list only shows languages whose model is installed; tap Download more languages to add one.
• Target (translation) – the language you want to read. Pick the same language as the source if you only want captions and no translation.

Swap flips the two. Auto-detect source lets the app guess the spoken language when you are not sure. The translation list comes from Google ML Kit (59 languages) or from your LibreTranslate server.

Use the searchable picker

Tapping either language opens a picker with a search box – type the first letters instead of scrolling. A lock with Pro next to a language means it is part of the Pro subscription; languages without the lock are included free.

Make the caption window readable

The Overlay card controls how the floating window looks:

• Text size and Opacity sliders.
• Show original text – on: the window shows two lines, what was said and the translation. Off: translation only.
• Overlay theme and Overlay font are Pro extras (Midnight Blue, Sunset, High Contrast; Rounded, Monospace, Serif).

You can change these while captioning is running; the window updates live.

Translation engine (the default is fine)

Further down, Translation engine decides who translates the text:

• On-device (ML Kit) – the default. Google ML Kit translates on the phone. The first time you use a language pair it downloads a ~30 MB model (you need internet once), then it works offline.
• LibreTranslate – point the app at your own LibreTranslate server for wider language coverage.

You never need to touch this for everyday use.

Server URLs only for self-hosters

If you choose LibreTranslate, a LibreTranslate URL field appears (example: http://<your-pc-ip>:5000). The app fetches /languages from it to fill the language lists; a red “Language fetch error” means the phone cannot reach that address – check the IP, the port and that both devices are on the same Wi-Fi. Remote Whisper has a matching Speech-to-Text URL field. The full Docker walkthrough is on the remote setup page.

Tap Start

Scroll back to the top and tap Start. The status changes to Listening, a persistent notification appears (Android requires one while the mic is in use) and the caption window shows up on screen.

Go anywhere – the window follows

Press Home, open YouTube, a meeting app, a browser – the caption window floats on top of all of them. Drag it by its bar to move it. The three buttons on it are Pause, Minimize and Close; the app remembers where you left it.

Watch the captions arrive

Words appear while they are being spoken, not after a pause. With Show original text on you get the spoken line on top and the translation underneath – here English speech translated into Vietnamese over a video playing in Google Photos. Partial lines update as the sentence completes.

Back in the app: live text and Stop

Return to LiveCaptionN at any time: the top card shows the current status and the latest line. Tap Stop when you are done; the window and the notification go away.

Everything is saved in History

The clock icon in the top bar opens Transcript History. Every session is stored on the phone with its language pair and time. Search it, Copy a line or Delete entries. Nothing is uploaded anywhere.

Help inside the app

The question-mark icon opens the built-in Tutorial & Guide with a narrated setup video, the hardware requirements and the self-hosted server steps. Report a Problem (bottom of the main screen) files a bug report on the app’s public GitHub issue tracker.