System naturalness prediction on FDX-Conv OOD (test)
0.362PCCConversational Speech Naturalness Predictor (Whisper)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Conversational Speech Naturalness Predictor (Whisper)Encoder=Whisper, Input Channel(s)=single2026.03 | 0.362 | 0.358 | 0.438 | |
| Conversational Speech Naturalness Predictor (AES)Encoder=AES, Input Channel(s)=single2026.03 | 0.31 | 0.297 | 0.548 | |
| Conversational Speech Naturalness Predictor (WavLM)Encoder=WavLM, Input Channel(s)=single2026.03 | 0.305 | 0.308 | 0.335 | |
| Conversational Speech Naturalness Predictor (Whisper)Encoder=Whisper, Input Channel(s)=dual2026.03 | 0.29 | 0.285 | 0.633 | |
| Conversational Speech Naturalness Predictor (AES)Encoder=AES, Input Channel(s)=dual2026.03 | 0.282 | 0.272 | 0.671 | |
| Conversational Speech Naturalness Predictor (WavLM)Encoder=WavLM, Input Channel(s)=dual2026.03 | 0.058 | 0.056 | 0.512 |