Paralinguistic-aware dialogue on ParaS2SBench Real (Human Evaluation)
4.911IEMOCAP ScoreGPT-TTS
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GPT-TTSCategory=Topline2025.11 | 4.911 | 4.925 | 4.918 | |
| Kimi-Audio GRPOCategory=Ours2025.11 | 4.336 | 3.859 | 4.098 | |
| Kimi-Audio SFTCategory=Ours2025.11 | 4.212 | 3.407 | 3.81 | |
| GPT-4o voice modeCategory=Closed Source2025.11 | 3.862 | 3.694 | 3.778 | |
| Qwen2.5 OmniCategory=Open Source2025.11 | 3.713 | 3.581 | 3.647 | |
| Whisper-GPT-TTSCategory=Baseline2025.11 | 3.601 | 3.552 | 3.487 | |
| GLM 4Category=Open Source2025.11 | 3.521 | 3.325 | 3.423 | |
| Kimi-AudioCategory=Open Source2025.11 | 2.231 | 2.272 | 2.252 |