Daedalus Echo

A standalone Android voice recorder, transcription, and AI analysis application. Records audio directly (using the built-in or Bluetooth microphone), transcribes recordings on-device with Whisper, generates smart summaries and interactive mind maps with Gemma 3, and lets you ask semantic questions across your entire library — all running 100% locally on-device.

Features

• Local Voice Recording — Record high-quality audio directly in the app using the device microphone or a Bluetooth headset (SCO)
• On-Device Transcription — Whisper base.en via sherpa-onnx; speech-to-text happens entirely offline and audio never leaves your phone
• AI Analysis — Gemma 3 1B generates a title, summary, key topics, and a structured mind map per recording
• Ask Your Notes — Semantic Q&A across your whole library: relevant recordings are retrieved by meaning, then Gemma synthesizes a detailed answer citing specific sources
• Knowledge Graph — Visualize semantic connections and shared topics across all of your recordings
• Full-Text Search — Instantly search across all transcripts and summaries
• Export — Share your summaries, mind maps, and Q&A answers as Markdown or copy them to the clipboard

Requirements

Hardware	Standard Android device with built-in microphone (Bluetooth mic/headset optional)
Android	ARM64, API 26+ (Android 8.0)
RAM	4 GB minimum (6 GB or more recommended for smooth execution)
GPU	Vulkan 1.1+ support recommended for GPU acceleration of Gemma 3
Storage	~770 MB free (Gemma 3 1B: ~555 MB · Whisper base.en: ~160 MB · text embedder: ~26 MB)

Device Compatibility

Since all processing is local, the app requires a modern processor and adequate memory. Generally, any mid-range or flagship Android device released in 2019 or later running Android 10+ with at least 4 GB of RAM is compatible.

Popular compatible device series include:

• Google Pixel: Pixel 3/3 XL/3a (minimum), Pixel 4 and newer (recommended).
• Samsung Galaxy:
  • S Series: S9, S10, S20, S21, S22, S23, S24, S25 and newer.
  • Note Series: Note 9, Note 10, Note 20 series.
  • A Series (Mid-range): A51, A52, A53, A54, A55, A71, A72, A73 and newer (6GB/8GB RAM models recommended).
  • Z Fold/Flip: All models.
• OnePlus: OnePlus 6T, 7, 8, 9, 10, 11, 12, 13 and newer.
• Xiaomi / Redmi / POCO:
  • Xiaomi: Mi 9, 10, 11, 12, 13, 14, 15 and newer.
  • Redmi Note: Note 8 Pro, 9 Pro, 10 Pro, 11 Pro, 12 Pro, 13 Pro and newer.
  • POCO: F3, F4, F5, F6, X3 Pro, X4 Pro, X5 Pro, X6 Pro.

Tested and verified on Samsung Galaxy S24 Ultra (Android 16, 12 GB RAM).

Setup

1. Open the android/ folder in Android Studio (Hedgehog or newer)
2. Build and run on your device, or via the command line:

   # Windows
   cd android && .\gradlew installDebug
   
   # macOS / Linux
   cd android && ./gradlew installDebug

3. On first launch the app will prompt you to download the AI models from Settings:
   • Gemma 3 1B (~555 MB) — on-device summarization and Q&A
   • Whisper base.en (~160 MB) — on-device speech-to-text
   • Universal Sentence Encoder (~26 MB) — on-device text embeddings for semantic Ask
4. Tap the record button on the home screen to start recording audio.

How It Works

All audio recording and AI processing happens locally on your Android device:

[ Built-in / BT Mic ] ──Record──► Android app
                                      │
                                 Whisper STT
                                 (on-device)
                                      │
                                 Gemma 3 1B
                                 (on-device)
                                      │
                           Title · Summary · Topics
                                 Mind Map

No data is sent to external servers.

Beyond per-recording analysis, Ask Your Notes embeds every transcript on-device so you can query the whole library by meaning — the most relevant recordings are retrieved, then Gemma composes an answer and links back to its sources.

Tech Stack

Component	Technology
UI	Jetpack Compose + Material 3
Architecture	MVVM · Room · StateFlow
On-device LLM	MediaPipe 0.10.35 + Gemma 3 1B (.task format)
Transcription	sherpa-onnx 1.13.2 + Whisper base.en int8 ONNX
Semantic search	MediaPipe Text Embedder + Universal Sentence Encoder
Audio Recording	AudioRecord + Bluetooth SCO support
Database	Room 2.6.1 + SQLite WAL
Build	AGP 8.7.3 · Kotlin 2.0.21 · JDK 21

Building

Requires Android SDK 35 and JDK 21.

cd android
.\gradlew assembleDebug    # debug APK
.\gradlew assembleRelease  # release APK (ADB debug hooks disabled)

Note

The first build downloads the prebuilt sherpa-onnx Android library (~56 MB) from GitHub Releases into app/libs/ automatically (the downloadSherpaOnnx Gradle task), so an internet connection is required for the initial build.

ADB broadcast commands (ANALYZE, SYNC, etc.) are only active in debug builds and are disabled in release builds.

License

MIT