AITranscribe Android

Native Android app for speech-to-text transcription with AI-powered post-processing. Designed for F-Droid distribution — all dependencies are FOSS, no Google Play Services required.

Features

• Push-to-Talk Recording — hold to record, release to transcribe; runs as a foreground service
• Multi-Provider STT — choose between Groq (whisper-large-v3-turbo, whisper-large-v3) and ZAI (GLM-ASR) for speech recognition
• LLM Post-Processing — automatically clean up or translate transcriptions via OpenRouter, Groq, or ZAI LLMs
• Background Transcription — recordings are transcribed via WorkManager; you can leave the app immediately
• Offline Queue — recordings are queued when offline and transcribed automatically when connectivity returns
• View Status Tracking — unviewed transcriptions are marked with a blue dot; auto-marked on open
• Search & Filter — filter by date range, text content, and view status (All / Only Unviewed)
• Export / Import — export transcription history to JSON or CSV; import from those formats
• Model Sync — model catalog is fetched periodically from each provider so you always see the latest available models
• Language Detection — detected language is stored per transcription; configure active languages per transcription
• Dark / Light Theme — follows system theme via Material 3

Screens

Screen	Purpose
Setup	First-launch wizard to enter Groq and OpenRouter API keys
Main	Recording button + list of recent transcriptions with search/settings toolbar
Detail	Full transcription text, post-processing actions (cleanup, translate), summary, share, delete
Search	Full-text search with date range and view filter
Settings	Provider management (auth/disconnect), STT/LLM model selection, language config, delete old transcriptions
Connect Provider	Add additional providers (Groq, OpenRouter, ZAI)

Architecture

Clean Architecture with three layers:

app/
├── data/           # Room database, Retrofit API clients, Repository implementations
│   ├── local/      # Room entities, DAOs, database, preferences
│   ├── remote/     # GroqApiService, OpenRouterApiService, ZaiApiService, NetworkModule
│   └── repository/ # TranscriptionRepository, AppPreferencesRepository
├── domain/         # Business logic (use cases), domain models, repository interfaces
│   ├── model/      # Transcription, ProviderConfig, ViewFilter, PostProcessingType, etc.
│   └── usecase/    # TranscribeAudioUseCase, PostProcessTextUseCase, SearchTranscriptionsUseCase, etc.
├── service/        # RecordingService (foreground), TranscriptionWorker (WorkManager), ModelSyncWorker
├── ui/             # Jetpack Compose screens, components, ViewModels, theme
│   ├── screen/     # MainActivity, MainScreen, SettingsScreen, SearchScreen, etc.
│   ├── components/ # BottomControlPanel, AudioRecordingButton, TranscriptionItem, etc.
│   ├── viewmodel/  # MainViewModel, SettingsViewModel, SearchViewModel, SetupViewModel
│   └── theme/      # Material 3 theme (Color, Theme, Type, Icon)
└── util/           # AudioChunker, NetworkMonitor, AppLogger, ToastManager

Key libraries: Jetpack Compose (Material 3), Room, Hilt, Retrofit + OkHttp, WorkManager, JUnit 5 + MockK + Turbine (testing).

Build

Prerequisites

• JDK 17+
• Android SDK 34
• Gradle (via wrapper)

Commands

# Debug build (faster, no minification)
./gradlew assembleDebug

# Release build (minified, obfuscated)
./gradlew assembleRelease

# Release AAB (for distribution)
./gradlew bundleRelease

# Install on connected device
./gradlew installDebug

APK output: app/build/outputs/apk/debug/app-debug.apk

Environment Variables for Tests

Copy .env.example to .env and fill in API keys for integration tests:

cp .env.example .env
# Edit .env with your keys

Configuration

API Keys

The app requires API keys for speech-to-text and (optionally) LLM post-processing:

Purpose	Provider	Key Format	Get One
STT	Groq	gsk_...	console.groq.com
STT	ZAI	API token	z.ai
LLM	OpenRouter	sk-or-...	openrouter.ai/keys
LLM	Groq	gsk_...	console.groq.com
LLM	ZAI	API token	z.ai

Enter keys in the Setup screen on first launch, or later in Settings → Connect Provider. Keys are stored encrypted on-device via AndroidX Security (EncryptedSharedPreferences).

Model Selection

In Settings you can choose separate providers and models for STT and LLM. Available models are synced from each provider's API and updated every 12 hours via ModelSyncWorker.

STT Models

Provider	Models
Groq	whisper-large-v3-turbo, whisper-large-v3
ZAI	glm-asr-2512

LLM Models (partial list)

Provider	Models
OpenRouter	inception/mercury, google/gemini-2.5-flash-lite, anthropic/claude-3-haiku, mistralai/mistral-small-3.1-24b-instruct, ...
Groq	llama-3.3-70b-versatile, llama-3.1-8b-instant, mixtral-8x7b-32768, gemma2-9b-it
ZAI	glm-4.7-flash, glm-4.5-flash, glm-4-32b-0414-128k, ...

The full model list is dynamic — synced from provider APIs on first connect and refreshed every 12 hours.

Build Config

• Min SDK: 30 (Android 11)
• Target SDK: 34 (Android 14)
• Debug build ID: com.georgernstgraf.aitranscribe.debug
• Release build ID: com.georgernstgraf.aitranscribe

Data Flow

Push & hold record button
    ↓
RecordingService (foreground) → MediaRecorder → .m4a file
    ↓  (release)
TranscriptionWorker (WorkManager)
    ├─ Groq / ZAI STT API → raw text + detected language
    └─ OpenRouter / Groq / ZAI LLM → cleaned text + summary
    ↓
Room database → transcription stored with status
    ↓
Notification: transcription complete

Testing

# Unit tests (JUnit 5 + MockK + Turbine)
./gradlew test

# Instrumented tests (requires device/emulator)
./gradlew connectedAndroidTest

Integration tests against live APIs are gated by the RUN_API_INTEGRATION_TESTS env flag.

License

FOSS — all dependencies are Apache 2.0, LGPL, or compatible. See individual library licenses.