Subjective Speech Evaluation Prediction on ConvTTS (val)
0.533Pearson RHuman Inter-Rater Consistency (Oracle)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Human Inter-Rater Consistency (Oracle)Input=Speech2026.02 | 0.533 | 0.517 | 0.301 | |
| GSRMInput=Speech2026.02 | 0.401 | 0.375 | 0.332 | |
| AES-based RegressorInput=Speech2026.02 | 0.384 | 0.352 | 0.228 | |
| WavLLM-based RegressorInput=Speech2026.02 | 0.331 | 0.279 | 0.242 | |
| Acoustic Feature Prompting (Gemini)Input=Text2026.02 | 0.133 | 0.13 | 1.18 | |
| Acoustic Feature Prompting (GPT-4o)Input=Text2026.02 | 0.1 | 0.103 | 0.549 | |
| Speech Prompting (Gemini)Input=Speech2026.02 | -0.05 | -0.078 | 1.009 | |
| NISQA (Min MOS)Input=Speech2026.02 | -0.104 | -0.109 | 3.325 | |
| UTMOSv2 (Min MOS)Input=Speech2026.02 | -0.161 | -0.165 | 6.19 | |
| NISQA (Mean MOS)Input=Speech2026.02 | -0.174 | -0.173 | 1.079 | |
| UTMOSv2 (Mean MOS)Input=Speech2026.02 | -0.191 | -0.206 | 0.987 |