Creatures Decoder — v2

Repo: Decode-Signals-of-Creatures-and-Plants
App name: Creatures Decoder

This is Version 2: adds a real FFT spectrogram (offline) and an ML classifier scaffold using TensorFlow Lite.

✅ What’s new in v2

• Spectrogram: decodes your recording to mono 16 kHz WAV via FFmpeg, computes FFT frames in Dart, and renders a grayscale heatmap.
• ML scaffold: screen to run inference via tflite_flutter. If no model is present, it fails gracefully with a placeholder result.

🔧 Dependencies

• ffmpeg_kit_flutter_min_gpl — for decoding AAC → WAV.
• tflite_flutter — TensorFlow Lite interpreter (no model bundled by default).
• flutter_sound, permission_handler, path_provider — from v1.

📂 Structure

lib/
 ├─ main.dart
 ├─ screens/
 │   ├─ home.dart            # Record/playback UI
 │   ├─ spectrogram.dart     # Real FFT spectrogram
 │   └─ classify.dart        # ML inference (graceful if model missing)
assets/
 └─ labels.txt               # Example labels (edit to match your model)

Place your TFLite model at assets/ml/model.tflite and update pubspec.yaml if you add the assets folder. In this starter, we only include labels.txt so the app builds without a model.

🏃 Run

flutter pub get
flutter create .    # if android/ios folders missing
flutter run

On iOS, add this to ios/Runner/Info.plist:

<key>NSMicrophoneUsageDescription</key>
<string>This app records audio to analyze animal/plant signals.</string>

🧠 Training & Models (next steps)

• Start with a small model that accepts 1-second mono audio @16 kHz (shape [1,16000]) and outputs class probabilities.
• Save as TFLite (.tflite) and include as assets/ml/model.tflite.
• Ensure assets/labels.txt has matching class order.

⚠️ Notes

• Spectrogram rendering is grayscale for simplicity. We can add better colormaps and axes later.
• The classifier will show a placeholder if the model is missing, so the UI flow remains testable.
• Plant ultrasonics need > 20 kHz capture; most phone mics won’t reach that. External ultrasonic mics are required.