Speaker Description on LibriTTS + DEMAND mixtures Background
99.5Gender AccuracyAAD-LLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AAD-LLMSystem Setup=Intention-Informed AAD-LLM2025.02 | 99.5 | 92.2 | 87.7 | |
| Qwen2-AudioSystem Setup=Upper Bound: Oracle Speaker + Auditory LLM2025.02 | 96.4 | 91.7 | 90 | |
| Qwen2-AudioSystem Setup=Proposed Baseline: Extracted Speaker + Auditory LLM2025.02 | 95.4 | 82.7 | 84.6 | |
| SALMONNSystem Setup=Auditory LLM without Intention2025.02 | 76.5 | 34.6 | 71.5 | |
| Qwen2-AudioSystem Setup=Auditory LLM without Intention2025.02 | 74.3 | 34.2 | 40.1 | |
| LTU-ASSystem Setup=Auditory LLM without Intention2025.02 | 73.1 | 46.5 | 28.1 | |
| Qwen2-AudioSystem Setup=Lower Bound: Random Speaker + Auditory LLM2025.02 | 72.6 | 63.1 | 73.8 | |
| GAMASystem Setup=Auditory LLM without Intention2025.02 | 58.8 | 34.5 | 50.5 | |
| WavLLMSystem Setup=Auditory LLM without Intention2025.02 | 54 | 33.7 | 51.2 | |
| Qwen-AudioSystem Setup=Auditory LLM without Intention2025.02 | 47.6 | 37.6 | 9 |