SpeakerSleuth - Discrimination Ranking: leaderboard
Ranking variant of the discrimination task in SpeakerSleuth (Lee, Pyo, Seo and Jo, Seoul National University, ACL 2026): the model orders the same three candidate clips by acoustic similarity to the target speaker, the reference order being the original, then the similar-voice conversion, then the opposite-gender conversion. The score is the percentage of the 606 items whose full order is exactly right (chance is one in six); higher is better.
Metric: Exact match of the full three-candidate ranking (%; 606 items). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 71.5 | #262 |
| 2 | Gemini 2.5 Flash | 61.6 | #452 |
| 3 | Gemini 2.5 Flash Lite | 23.1 | #958 |
| 4 | Qwen2.5-Omni-7B | 22.9 | #1797 |
| 5 | Phi-4 Multimodal Instruct | 21.3 | #3914 |
| 6 | Gemma 3n E4B | 20.1 | #3330 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=speakersleuth-discrimination-ranking · How It Works · Data refreshed daily, snapshot 2026-11-12.