Subjective Speech Evaluation Prediction on FDX-Conv (out-of-domain)
0.532Pearson CorrelationHuman Inter-Rater Consistency (Oracle)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Human Inter-Rater Consistency (Oracle)Input=Speech2026.02 | 0.532 | 0.489 | 0.336 | |
| GSRMInput=Speech2026.02 | 0.465 | 0.427 | 0.23 | |
| Acoustic Feature Prompting (GPT-4o)Input=Text2026.02 | 0.277 | 0.214 | 0.317 | |
| UTMOSv2 (Min MOS)Input=Speech2026.02 | 0.245 | 0.244 | 2.815 | |
| AES-based RegressorInput=Speech2026.02 | 0.236 | 0.224 | 0.265 | |
| NISQA (Min MOS)Input=Speech2026.02 | 0.229 | 0.212 | 1.066 | |
| WavLLM-based RegressorInput=Speech2026.02 | 0.222 | 0.212 | 0.275 | |
| UTMOSv2 (Mean MOS)Input=Speech2026.02 | 0.196 | 0.21 | 0.442 | |
| Acoustic Feature Prompting (Gemini)Input=Text2026.02 | 0.193 | 0.157 | 0.814 | |
| NISQA (Mean MOS)Input=Speech2026.02 | 0.164 | 0.147 | 1.197 | |
| Speech Prompting (Gemini)Input=Speech2026.02 | 0.14 | 0.136 | 0.531 |