Ablation Study Results for FedEmoNet Speech Emotion Recognition Framework

Mohammed Tawfik published results from ablation experiments for the FedEmoNet framework on 2026-05-07. The dataset, 5.5 KB in size, contains results from a federated learning study on speech emotion recognition using German (EmoDB) and English (RAVDESS) speech corpora. It includes performance metrics for model components like PSO feature selection, Transformer blocks, and the FedProx protocol.

Use Cases

• Evaluating the impact of PSO feature selection on model accuracy based on the reported +2.80% contribution.
• Analyzing the performance contribution of Transformer blocks in a hybrid TCN-Transformer architecture based on the reported +2.10% gain.
• Assessing the effect of the FedProx protocol in federated optimization based on the reported +2.62% improvement.
• Studying cross-corpus generalization patterns for speech emotion recognition based on the reported zero-shot evaluation results on CREMA-D.

Strengths

• Results include specific accuracy improvements for key components: PSO selection (+2.80%), Transformer blocks (+2.10%), and FedProx protocol (+2.62%).
• Cross-corpus evaluation provides zero-shot accuracy of 68.15% ± 1.23% on CREMA-D, with breakdowns for high-arousal (71.9%) and low-arousal (62.1%) emotions.
• Framework achieved high accuracy on held-out test sets: 99.07% ± 0.35% on EmoDB and 98.96% ± 0.42% on RAVDESS.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small (5.5 KB), indicating limited scope, likely containing summary results rather than raw data.

Provenance

Source

figshare

Collection Method

Results from ablation experiments for the FedEmoNet federated learning framework.

Freshness

Last updated 2026-05-07 17:45:48; freshness should be verified.