Speaker Consistency Alignment Prediction on Mandarin Gold (test)
0.759LCCTTS-PRISM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TTS-PRISMParameters=7B, Inference Strategy=single-pass inference2026.04 | 0.759 | 0.752 | 0.022 | |
| Gemini-2.5-ProInference Strategy=dimension-wise inference2026.04 | 0.733 | 0.718 | 0.029 | |
| Step-Audio-R1Parameters=33B, Inference Strategy=dimension-wise inference2026.04 | 0.591 | 0.581 | 0.064 | |
| Qwen3-OmniParameters=30B, Inference Strategy=dimension-wise inference2026.04 | 0.582 | 0.563 | 0.041 |