Agitation prediction on Bridge2AI-Voice 5-fold speaker-independent CV v3.0.0
0.117Pearson Correlation (ρ)MP-IB
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MP-IBParams=657K, Size (KB)=6172026.05 | 0.117 | 1.05 | |
| MI-Minimization (MINE)Params=520K, Size (KB)=1,0402026.05 | 0.095 | 1.08 | |
| β-VAE DisentanglementParams=450K, Size (KB)=9002026.05 | 0.089 | 1.12 | |
| On-device-ERParams=280K, Size (KB)=5602026.05 | 0.081 | 1.15 | |
| GRL (Same Backbone)Params=657K, Size (KB)=1,1402026.05 | 0.079 | 1.14 | |
| TinyML-SERParams=120K, Size (KB)=2402026.05 | 0.078 | 1.22 | |
| VQ (Same Backbone)Params=657K, Size (KB)=1,1402026.05 | 0.074 | 1.16 | |
| Adversarial-MLPParams=45K, Size (KB)=902026.05 | 0.072 | 1.18 | |
| VQ-DisentanglementParams=85K, Size (KB)=1702026.05 | 0.069 | 1.21 | |
| Uniform INT4 SERParams=32K, Size (KB)=162026.05 | 0.061 | 1.25 | |
| Shallow CNN (120K)Params=120K, Size (KB)=2402026.05 | 0.058 | 1.32 | |
| WavLM-Adapter (Sup.)Params=94.8M, Size (KB)=189,6002026.05 | 0.042 | 0.95 | |
| Hand-Crafted Prosody (HCP)2026.05 | 0.031 | 1.45 | |
| ECAPA-TDNN-Adapter (Sup.)Params=6.7M, Size (KB)=13,4002026.05 | 0.031 | 0.98 | |
| WavLM-Adapter+SSL (Sup.)Params=94.8M, Size (KB)=189,6002026.05 | 0.031 | 0.92 | |
| ECAPA-TDNN+Leakage Supp.Params=6.7M, Size (KB)=13,4002026.05 | 0.025 | 0.99 |