Offline Voice Input

notune

About this app

Offline Voice Input delivers professional-grade speech-to-text that runs entirely on your device. Whether you're writing emails, taking notes, or chatting, your voice never leaves your phone — no internet connection, no account, no tracking.

VOICE TYPING IN ANY APP
Tap the microphone on a compatible keyboard — such as Microsoft SwiftKey — or a website's voice-search button, and Offline Voice Input transcribes your speech straight into the text field. It registers as your device's speech-to-text provider.
Note: some keyboards (such as Gboard) only use Google's own voice typing and can't pass speech to other apps. If yours does, switch to a keyboard like SwiftKey — or simply use the built-in voice keyboard described below, which always works.

LIVE SUBTITLES
Get real-time, on-device captions for any audio or video playing on your phone — videos, podcasts, meetings and more.

TRANSCRIBE AUDIO FILES
Share an audio file to Offline Voice Input (or open it with the app) to turn the recording into text.

OPTIONAL VOICE KEYBOARD
Prefer voice input as its own keyboard? A dedicated voice keyboard is included — switch to it whenever you like.

WHY YOU'LL LOVE IT
• 100% offline — even works in airplane mode, nothing is ever uploaded
• Private by design — audio is processed on-device and never stored or sent
• Fast and accurate, powered by NVIDIA's Parakeet model
• 25 languages, including English, German, French, Spanish, Italian, Portuguese, Dutch, Polish, Russian and Ukrainian
• Free and open source (MIT License)

The app downloads the speech model (~670 MB) once on first launch; afterwards everything runs locally on your device.

POWERED BY OPEN SOURCE
We believe in transparency and the power of the community. Offline Voice Input is built with cutting-edge open-source technology:
• NVIDIA Parakeet TDT 0.6B — high-performance on-device speech recognition (INT8 ONNX, quantized by istupakov)
• transcribe-rs — the Rust transcription engine by CJ Pais
• ONNX Runtime — efficient on-device model inference
• A native Android interface with a Rust core for speed and safety

Source code, issues and contributions:
https://github.com/notune/android_transcribe_app

Ratings and reviews

Multiple issues. Can't see what you're saying when speaking. You have to tap stop then wait as it processes and outputs everything at once. Sometimes it won't process what you said at all or outputs only a couple of words. It is a transformer model so instead of hearing what you say and writing, it outputs what it thinks you said based on what is commonly said. One word said outputs random phrases. Capitalizes words in the middle of sentences. Overly punctuates.

Even faster than Futo. Parakeet V3 is an amazing model. Would it be possible to have the keyboard designated as voice input only so that my regular keyboard pops up first? By the way, for the time it took to put this app together, it's worth a couple dollars. A tip to fix one of the annoying problems in the app. The bottom right hand corner is the systems icon to switch to another keyboard. That's a lot easier to tap than the icon within the app.

Great app. The only thing that's bothering me is that I need to select Offline Voice Input each time I use the vocal typing because only the Xiaomi option has the "always" / "once" option... Pretty annoying but I don't think it comes from the app itself. I'm french and french recognition works fine, some weird translation appears time to time but hey... Haha. Anyway, it's free, its doing the job, this is great. Thanks to the devs! 🫡

What’s new

• New engine: transcribe.cpp – model now just 485 MB, faster via SIMD
• Import your own GGUF models fully offline
• Language picker & translate-to-English option
• Thread setting with built-in benchmark
• Draggable live subtitles
• Optionally skip the screen-capture prompt via adb
• More stable: no more stuck "Processing" or endless retries
• Now in DE, FR, ES, IT, PT, RU
• UI polish & clearer keyboard help