Speaker Profiling on VoxCeleb1-O
99.9Gender AccuracyAudio Flamingo3
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Audio Flamingo3LLM Backbone=Qwen2.5-7B2026.05 | 99.9 | 23.2 | 59.8 | 22.1 | 38.3 | |
| SpeakerLLM-BaseLLM Backbone=Qwen2.5-1.5B2026.05 | 99.9 | 39.8 | 83.1 | 72.4 | 54.2 | |
| Qwen2.5-Omni-7BLLM Backbone=Qwen2.5-7B2026.05 | 99.8 | 17.5 | 76 | 22.7 | 25 | |
| Qwen3.0-Omni-30BLLM Backbone=Qwen3-30B-A3B2026.05 | 99.1 | 20.3 | 75.7 | 23.8 | 32.1 | |
| CoLMboLLM Backbone=GPT2-base2026.05 | 78.6 | 22.7 | 45.5 | — | — |