FedEmoNet: Federated Speech Emotion Recognition Results

FedEmoNet achieves 99.07% accuracy on EmoDB and 98.96% on RAVDESS using federated learning with differential privacy. The framework, authored by Mohammed Tawfik, demonstrates cross-corpus generalization with 68.15% accuracy on CREMA-D. Results were last updated in May 2026.

Use Cases

• Benchmark federated learning performance based on reported accuracy metrics
• Analyze cross-corpus generalization patterns based on arousal-dependent transfer results
• Evaluate privacy guarantees based on membership inference attack resistance metrics
• Study feature importance for emotion recognition based on SHAP-LIME explainability analysis

Strengths

• High accuracy reported on two primary datasets (99.07% on EmoDB, 98.96% on RAVDESS)
• Cross-corpus evaluation on CREMA-D yields 68.15% accuracy
• Framework includes formal differential privacy guarantees with attack resistance AUC of 0.52
• Ablation experiments quantify contributions of specific components (PSO +2.80%, Transformer +2.10%, FedProx +2.62%)

Limitations

• Dataset size is 5.5 KB, indicating a very limited scope of results
• Row count is unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download

Provenance

Source

figshare

Collection Method

Results from federated learning experiments combining FedProx, TCN-Transformer, PSO feature selection, and differential privacy.

Freshness

Last updated 2026-05-07 17:45:49; freshness should be verified