Subjective Speech Evaluation on Sarcastic Speech Synthesis Dataset
3.8NMOSGround Truth
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ground TruthSemantic Cue Type=-, Prosodic Cue Type=-2025.10 | 3.8 | 4.5 | |
| w/ LoRA + RAGSemantic Cue Type=✓, Prosodic Cue Type=✓2025.10 | 2.7 | 3.8 | |
| BaselineSemantic Cue Type=-, Prosodic Cue Type=-2025.10 | 2.6 | 3.2 | |
| w/ LLaMA 3-LoRASemantic Cue Type=✓, Prosodic Cue Type=-2025.10 | 2.6 | 3.8 | |
| w/ RAGSemantic Cue Type=-, Prosodic Cue Type=✓2025.10 | 2.6 | 3.7 | |
| w/ BERTSemantic Cue Type=✓, Prosodic Cue Type=-2025.10 | 2.5 | 3.1 | |
| w/ LLaMA 3Semantic Cue Type=✓, Prosodic Cue Type=-2025.10 | 2 | 2.6 |