synheart_emotion 0.1.0

On-device emotion inference from biosignals (HR/RR) for Flutter applications. Infers momentary emotions (Amused, Calm, Stressed) from heart rate and RR intervals using WESAD-trained machine learning models.

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

0.1.0 - 2025-01-28

Added

• Initial release of synheart-emotion library
• 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
• JSON model loader for loading trained models from assets
• Comprehensive test suite with unit tests and performance benchmarks (31 tests)
• Example app demonstrating integration with wearable devices
• JSON model example showing asset-based model loading
• Complete documentation (RFC-E1.1, MODEL_CARD, README, CONTRIBUTING)
• GitHub Actions CI/CD pipeline
• Strict linting with analysis_options.yaml

Changed

• 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 real model usage and accuracy metrics

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
• Requires minimum 30 RR intervals (configurable) for inference
• JSON model loading requires Flutter asset bundling

Dependencies

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

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