System-wide voice dictation for Android built around Whisper Large v3 instead of Gboard’s real-time engine — much better accuracy on long-form dictation. Tap the floating bubble to start, tap again to stop, and the transcript lands at your cursor in whatever app you’re in.
Requires Android 12 (API 31) or later • BYOK (free tier on Groq) • install instructions
Accessibility-Service apps need a separate Play Store policy review, so for now SrizonVoice ships as a sideload from GitHub Releases. The APK is signed with a stable release key — you can install future updates in-place without uninstalling.
Android will warn you about installing from an unknown source; that’s normal for sideloaded apps. See the install instructions below for the two-tap walkthrough.
Android already has voice typing built into the keyboard. It’s fine for one-sentence replies, and falls apart on anything longer. Here’s the gap SrizonVoice is filling.
Google’s built-in voice typing transcribes as you speak, in real time. It optimizes for latency, so on long recordings it drops chunks, restarts mid-sentence, garbles proper nouns, and never sees the full sentence in context. Good enough for a quick reply, painful for a paragraph.
Records the whole take first, then sends it to OpenAI’s Whisper Large v3 in one shot. The model gets every word in context — so punctuation, paragraphs, technical terms, code, and names all hold up. Optional Gemini pass on top to clean up filler words.
The tradeoff: you wait a second or two after releasing instead of watching words appear live. For anything longer than a sentence, it’s worth it.
Same Groq Whisper pipeline as the macOS app, adapted for Android idioms
Tap the floating bubble once to start, tap again to stop. Default mode — better for long dictations than holding a finger on a tiny target. Optional auto-stop on silence.
Prefer the macOS-style hold? Flip a setting to switch the bubble to push-to-talk: hold to record, release to transcribe, drag up to cancel.
An Accessibility Service inserts your transcript directly into the focused text field. Falls back to clipboard paste when the field doesn’t expose proper accessibility nodes.
Optional Gemini pass cleans up filler words and punctuation without changing meaning — same prompt as the macOS app.
Dictate in any of 107 supported languages. Recently-used languages appear at the top of the picker for quick switching.
30-bar live waveform with the same coral → purple → blue gradient as the macOS app, driven by RMS at 60 Hz.
Your Groq (and optional Gemini) API key is stored in EncryptedSharedPreferences. Audio is sent directly to Groq — no Srizon servers in the middle.
Three steps to hands-free typing, from any Android app
Tap the floating bubble from any app to start recording. Hold instead if you've switched to push-to-talk mode.
A live 30-bar waveform reacts to your voice. Audio is captured at 16 kHz mono — Whisper’s native rate.
Tap once more (or auto-stop on silence). Whisper transcribes, Gemini optionally cleans up, and the Accessibility Service inserts text at your cursor.
Two paths: grab the prebuilt APK (recommended), or build it yourself from source
Option A · Recommended
Grab srizonvoice-1.1.0.apk (~46 MB) directly from GitHub Releases. You can download it on your phone, or on a computer and transfer it across.
Open Settings → Apps → Special access → Install unknown apps on your phone, pick the app you used to download the APK (Chrome, Files, Drive, etc.), and toggle Allow from this source.
Tap the downloaded file. Android will show a confirmation sheet — review it and tap Install. Already on an earlier SrizonVoice release? This installs as an in-place upgrade (same signing key), so no uninstall needed.
Paste a Groq key (free tier at console.groq.com), optionally a Gemini key, then grant Microphone, Notifications, Display-over-other-apps, and Accessibility permissions when prompted.
Option B · For developers
Clone github.com/AfzalH/voice-android or download the source as a ZIP.
Android Studio Ladybug (2024.2.1) or later. It bundles JDK 17 and Gradle 8.10 and will generate the missing gradle-wrapper.jar on first sync.
Run ./gradlew assembleDebug — or use Build → Build Bundle(s) / APK(s) → Build APK(s) in the IDE.
Enable Developer options → USB debugging, plug in a device running Android 12+ (API 31+), and run adb install app/build/outputs/apk/debug/app-debug.apk. Or just hit Run in Android Studio.
Heads up:debug builds are signed with Android’s debug key, so they conflict with the release APK. You’ll need to uninstall one before installing the other.
Audio is only captured during an active dictation session — from the tap that starts recording to the tap (or auto-stop) that ends it — and is sent directly from your device to Groq for transcription. If you enable post-processing, the transcript is then sent to Google Gemini. No data passes through Srizon servers, and your API key is stored in EncryptedSharedPreferences.
See Groq Privacy Policy and Google Privacy Policy for AI provider policies.
Download the APK, grant a few permissions, paste a Groq key — you’ll be dictating into any Android app inside two minutes.