AI Subtitles Offline: Generate Subtitles From Any Video on Your Phone

Some videos simply have no subtitles: a regional film, an old family recording, a lecture, a clip someone forwarded on WhatsApp. Searching for a matching .srt file goes nowhere. VideoPlayer can now listen to the video and write AI subtitles for it — on your phone, without uploading anything, even with the internet switched off.

How AI subtitles work

VideoPlayer uses Whisper, an open-source speech-recognition model, running through the efficient whisper.cpp engine. The player decodes the video’s sound, cuts it into pieces of about 30 seconds at natural pauses, and the model turns each piece into text with timestamps. Those become subtitle lines on screen.

The work happens a little ahead of where you’re watching, so the subtitles are ready when the scene arrives. It runs in a separate process, so even on a phone that’s low on memory, the video keeps playing smoothly.

How to turn on AI subtitles

1. Play the video and tap the CC button.
2. Under Sync & tools, tap AI subtitles.
3. Choose a model the first time (see below) and the language spoken in the video — or leave it on Auto.
4. Tap Download & start. After the one-time download it just says Start.

Subtitles appear within the first half minute and keep coming as you watch. Jump far ahead and the AI starts again from the new spot.

Which AI subtitles model should you pick?

You can delete the model any time from the same screen to free the space.

AI subtitles in English for any language

Switch on Make English subtitles and the AI translates while it listens — a Japanese anime, a Korean drama or a Tamil film comes out with English lines. To get subtitles in Hindi instead, keep the original language and use Translate subtitles on top of the AI subtitles.

How accurate are AI subtitles?

Clear speech with little background music comes out remarkably well. Accuracy drops with heavy music, several people talking over each other, strong slang, or very noisy recordings. A few honest notes:

• Names and places are sometimes spelled the way they sound.
• Songs are mostly skipped — the model is tuned for speech.
• Hindi output is written in Devanagari. Hinglish speech may come out mixed.
• Older phones may run slower than the video; the subtitles then appear a bit later.

If a video already has a subtitle file that’s just out of time, auto sync is quicker and more accurate than generating new ones.

Privacy: nothing leaves your phone

The model file is downloaded once from Hugging Face. After that, the sound of your video is processed on the phone only. No account, no upload, no server — which is why AI subtitles work on a plane, in a village with no signal, and for videos you’d never want to send anywhere.

To learn more about the model itself, see the whisper.cpp project on GitHub.

Frequently asked questions