On-device voice transcription — no cloud, no API, everything offline.
Record voice notes and transcribe them directly on your Android device using Whisper.cpp. Your voice never leaves your phone.
- 🎙️ Record & Transcribe — one tap to capture audio, automatic transcription
- 🔒 100% Local — Whisper Small (466 MB) runs on-device after first download, zero network requests for transcription
- 🌍 Multilingual — Russian, English, and 97+ languages via Whisper models
- ⚡ Quality — Whisper Small model for high accuracy speech recognition across 99 languages
- 📱 Android 8+ — works on devices from 2017 onward
- 🎨 Material Design — clean, simple UI
Current version: 1.0.0 (May 2026)
# Clone
git clone https://github.com/570775-ux/Benetto.git
cd Benetto
# Build (requires Android SDK 36+)
export ANDROID_SDK_ROOT=~/android-sdk
./gradlew assembleRelease
# APK at app/build/outputs/apk/release/app-release.apk┌─────────────────────────────────┐
│ VoiceNotes Local (Android) │
│ ┌───────────┐ ┌─────────────┐ │
│ │ Recording │ │ Transcription│ │
│ │ Module │→ │ Engine │ │
│ │ (AudioRec)│ │ (Whisper.cpp)│ │
│ └───────────┘ └─────────────┘ │
│ ↓ ↓ │
│ ┌─────────────────────────────┐│
│ │ Room Database ││
│ │ (notes, transcripts, meta) ││
│ └─────────────────────────────┘│
│ ↓ │
│ ┌─────────────────────────────┐│
│ │ Jetpack Compose UI ││
│ └─────────────────────────────┘│
└─────────────────────────────────┘
- Audio Processing: Custom WAV encoder, optimized for Whisper input format (16kHz, mono, 16-bit PCM)
- Transcription Engine: Whisper.cpp via JNI, Small model (~500MB), ~2-3x real-time
- Data Layer: Room + Repository pattern, offline-first
- UI: Jetpack Compose, Material 3
| Layer | Technology |
|---|---|
| UI | Jetpack Compose, Material 3 |
| Architecture | MVVM + Repository Pattern |
| Database | Room (SQLite) |
| Audio | Android MediaRecorder + Custom WAV processor |
| Transcription | Whisper.cpp (C++ via JNI) |
| DI | Manual (lightweight, no framework) |
| Build | Gradle (Kotlin DSL) |
See ROADMAP.md for planned features and milestones.
See CONTRIBUTING.md for guidelines.
MIT License — see LICENSE for details.
Built with ❤️ for privacy. No cloud. No tracking. Just your voice, your device.