Breaking Changes #

• Removed LinearSvmModel class: The LinearSvmModel class and model_linear_svm.dart file have been removed. The package now focuses exclusively on ONNX models.
• Removed json_linear_model.dart export: The dead export for json_linear_model.dart has been removed (file was missing).
• Package structure change: Moved lib/ directory from repository root to sdks/flutter/lib/ to follow proper Flutter package conventions.

Migration Guide #

If you were using LinearSvmModel, migrate to OnnxEmotionModel:

// Old (0.1.0)
import 'package:synheart_emotion/synheart_emotion.dart';

final model = LinearSvmModel.fromArrays(
  modelId: 'wesad_emotion_v1_0',
  version: '1.0',
  labels: ['Amused', 'Calm', 'Stressed'],
  featureNames: ['hr_mean', 'sdnn', 'rmssd'],
  weights: [...],
  biases: [...],
  mu: {...},
  sigma: {...},
);

final engine = EmotionEngine.fromPretrained(
  config,
  model: model,
);

// New (0.2.0)
import 'package:synheart_emotion/synheart_emotion.dart';

final model = await OnnxEmotionModel.loadFromAsset(
  modelAssetPath: 'assets/ml/extratrees_wrist_all_v1_0.onnx',
  metaAssetPath: 'assets/ml/extratrees_wrist_all_v1_0.meta.json',
);

final engine = EmotionEngine.fromPretrained(
  config,
  model: model,
);

Fixed #

• Fixed Flutter package structure: lib/ directory is now properly located in sdks/flutter/lib/ where pubspec.yaml is located, enabling proper pub.dev publishing.

Changed #

• Package now exclusively uses ONNX models for emotion inference.
• Improved package structure alignment with Flutter package conventions.

This release includes synchronized versions across all platforms. All SDKs are at version 0.1.0.

Added #

Multi-Platform SDK Support

• Flutter/Dart SDK: Primary SDK for Flutter applications

  • Published to pub.dev as synheart_emotion
  • Full API parity with all features
  • Asset-based model loading support

• Python SDK: Cross-platform Python package

  • Published to PyPI as synheart-emotion
  • Installable via pip install synheart-emotion
  • Complete feature implementation with NumPy/Pandas

• Android SDK: Kotlin library for Android applications

  • Published to Maven Central as ai.synheart:emotion
  • Minimum SDK 21 (Android 5.0+)
  • Coroutines support for async operations

• iOS SDK: Swift package for iOS/macOS/watchOS/tvOS

  • Available via Swift Package Manager
  • Published to CocoaPods as SynheartEmotion
  • Supports iOS 13.0+, macOS 10.15+, watchOS 6.0+, tvOS 13.0+

Core Features

• Core emotion inference engine with sliding window buffering
• Support for three emotion states: Amused, Calm, Stressed
• Feature extraction from heart rate and RR intervals (HR mean, SDNN, RMSSD)
• Real WESAD-trained Linear SVM model with 78% accuracy
• One-vs-rest classification with softmax probability output
• Configurable inference parameters (window size, step size, min RR count)
• Real-time streaming API via EmotionStream (Flutter)
• JSON model loader for loading trained models from assets
• Comprehensive test suite with unit tests and performance benchmarks
• Example applications for each platform
• Complete documentation (RFC-E1.1, MODEL_CARD, README, CONTRIBUTING)

Publishing Infrastructure

• GitHub Actions workflows for automated publishing:
  • Flutter → pub.dev (OIDC trusted publishing)
  • Python → PyPI (OIDC trusted publishing)
  • Android → Maven Central (with GPG signing)
  • iOS → Swift Package Manager + CocoaPods
• Automated CI/CD pipeline with tests and validation
• Version management across all SDKs
• Publishing documentation and guides

Changed #

• Repository structure: Restructured into monorepo with separate SDKs and examples directories
• Model weights: Replaced placeholder weights with real WESAD-trained parameters
• Model ID: Updated from svm_linear_wrist_sdnn_v1_0_PLACEHOLDER to wesad_emotion_v1_0
• Model version: Updated from 1.0-demo to 1.0
• Normalization parameters: Updated to WESAD dataset statistics
• Documentation: Updated README to reflect multi-platform SDKs, publishing setup, and real model usage

Architectural Decisions #

• No built-in storage: Library focuses on emotion inference only
  • Users can implement their own storage using preferred methods
  • Reduces dependencies and security concerns
  • Provides flexibility for different storage requirements
• Privacy-first: All processing happens on-device, no network calls
• Minimal dependencies: Only requires Flutter SDK
• Asset-based models: Support for loading JSON models from Flutter assets
• SWIP SDK compatibility: Follows SWIP SDK model format and architecture patterns

Security Notes #

• ✅ Real trained model: Default SVM weights are trained on WESAD dataset
• Production ready: Model achieves 78% accuracy on validation data
• Data storage: Not included - implement your own secure storage as needed

Model Performance #

• Dataset: WESAD (15 subjects, 1200 windows)
• Accuracy: 78% on 3-class emotion recognition
• Balanced Accuracy: 76%
• F1-Score: 75%
• Features: HR mean, SDNN, RMSSD
• Classes: Amused, Calm, Stressed

Known Limitations #

• Limited to three emotion classes (Amused, Calm, Stressed)
• Requires minimum 30 RR intervals (configurable) for inference
• JSON model loading requires platform-specific asset bundling
• Android: Manual staging repository release required after publishing workflow

Dependencies #

Flutter SDK

• Flutter >=3.10.0
• Dart SDK >=3.0.0 <4.0.0
• No external package dependencies (beyond Flutter SDK)

Python SDK

• Python >=3.8
• NumPy >=1.21.0
• Pandas >=1.3.0
• Optional: scikit-learn, joblib, xgboost (for ML features)

Android SDK

• Android API 21+ (Android 5.0 Lollipop)
• Kotlin 1.8+
• AndroidX Core KTX
• Kotlin Coroutines

iOS SDK

• iOS 13.0+ / macOS 10.15+ / watchOS 6.0+ / tvOS 13.0+
• Swift 5.0+
• No external dependencies

Performance #

• HR mean calculation: 0.008ms average
• SDNN calculation: 0.007ms average
• RMSSD calculation: 0.002ms average
• Full feature extraction: 0.006ms average
• Model inference: 0.008ms average
• Full inference cycle: 0.010ms average
• All metrics well under <5ms target latency (measured on mid-range devices)

Installation #

Flutter

dependencies:
  synheart_emotion: ^0.1.0

Python

pip install synheart-emotion

Android

dependencies {
    implementation("ai.synheart:emotion:0.1.0")
}

iOS

Swift Package Manager:

dependencies: [
    .package(url: "https://github.com/synheart-ai/synheart-emotion.git", from: "0.1.0")
]

CocoaPods:

pod 'SynheartEmotion', '~> 0.1.0'

Publishing Information #

• Flutter: Published to pub.dev
• Python: Published to PyPI
• Android: Published to Maven Central
• iOS: Available via Swift Package Manager and CocoaPods

Acknowledgments #

• WESAD dataset for training data
• All contributors and testers