Paralinguistic-aware dialogue on ParaS2SBench Synthetic (Human Evaluation)
4.922Average Paralinguistic Awareness ScoreGPT-TTS
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GPT-TTSCategory=Topline2025.11 | 4.922 | — | — | — | — | |
| GPT-TTSCategory=Topline2025.11 | 4.844 | 4.752 | 4.889 | 4.923 | 4.813 | |
| Kimi-Audio GRPOCategory=Ours2025.11 | 4.407 | 4.316 | 4.51 | 4.381 | 4.422 | |
| Kimi-Audio GRPOCategory=Ours2025.11 | 4.303 | — | — | — | — | |
| Kimi-Audio SFTCategory=Ours2025.11 | 4.089 | 4.192 | 4.223 | 3.812 | 4.131 | |
| Kimi-Audio SFTCategory=Ours2025.11 | 3.996 | — | — | — | — | |
| GPT-4o voice modeCategory=Closed Source2025.11 | 3.564 | — | — | — | — | |
| Qwen2.5 OmniCategory=Open Source2025.11 | 3.538 | — | — | — | — | |
| Qwen2.5 OmniCategory=Open Source2025.11 | 3.483 | 3.352 | 3.953 | 3.496 | 3.131 | |
| GPT-4o voice modeCategory=Closed Source2025.11 | 3.457 | 3.375 | 3.833 | 3.542 | 3.078 | |
| Whisper-GPT-TTSCategory=Baseline2025.11 | 3.23 | — | — | — | — | |
| GLM 4Category=Open Source2025.11 | 3.23 | — | — | — | — | |
| GLM 4Category=Open Source2025.11 | 3.134 | 3.012 | 3.514 | 3.228 | 2.781 | |
| Whisper-GPT-TTSCategory=Baseline2025.11 | 3.102 | 3.212 | 3.041 | 3.042 | 3.112 | |
| Kimi-AudioCategory=Open Source2025.11 | 2.924 | 3.278 | 2.382 | 3.121 | 2.912 | |
| Kimi-AudioCategory=Open Source2025.11 | 2.699 | — | — | — | — |