Speaker drift detection on Speaker drift detection dataset 128 samples (full)
90.7F1 ScoreGPT-4o
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4oLLM Backbone=GPT-4o, Embedding Type=Wav2Vec2, Thresholds=(0.960, 0.950, 0.995)2026.04 | 90.7 | |
| DeepSeek-R1LLM Backbone=DeepSeek-R1, Embedding Type=Whisper, Thresholds=(0.970, 0.990, 0.995)2026.04 | 90.6 | |
| GPT-4oLLM Backbone=GPT-4o, Embedding Type=Whisper, Thresholds=(0.960, 0.950, 0.995)2026.04 | 89.41 | |
| Claude-4LLM Backbone=Claude-4, Embedding Type=Wav2Vec2, Thresholds=(0.950, 0.950, 0.995)2026.04 | 88.2 | |
| Claude-4LLM Backbone=Claude-4, Embedding Type=Whisper, Thresholds=(0.950, 0.950, 0.995)2026.04 | 84.38 | |
| Gemini-Pro-2.5LLM Backbone=Gemini-Pro-2.5, Embedding Type=Wav2Vec2, Thresholds=(0.970, 0.950, 0.998)2026.04 | 82.9 | |
| Gemini-Pro-2.5LLM Backbone=Gemini-Pro-2.5, Embedding Type=Whisper, Thresholds=(0.970, 0.950, 0.998)2026.04 | 80 | |
| DeepSeek-R1LLM Backbone=DeepSeek-R1, Embedding Type=Wav2Vec2, Thresholds=(0.970, 0.990, 0.995)2026.04 | 80 | |
| DeepSeek-R1LLM Backbone=DeepSeek-R1, Embedding Type=MFCC, Thresholds=(0.970, 0.990, 0.995)2026.04 | 80 | |
| Claude-4LLM Backbone=Claude-4, Embedding Type=MFCC, Thresholds=(0.950, 0.950, 0.995)2026.04 | 76 | |
| Qwen-3LLM Backbone=Qwen-3, Embedding Type=Wav2Vec2, Thresholds=(0.910, 0.950, 0.990)2026.04 | 72.7 | |
| Qwen-3LLM Backbone=Qwen-3, Embedding Type=Whisper, Thresholds=(0.910, 0.950, 0.990)2026.04 | 72.7 | |
| Qwen-3LLM Backbone=Qwen-3, Embedding Type=MFCC, Thresholds=(0.910, 0.950, 0.990)2026.04 | 71.4 | |
| PCA-based BaselineEmbedding Type=Wav2Vec2, Thresholds=(0.950, 0.950, 0.950)2026.04 | 71.3 | |
| PCA-based BaselineEmbedding Type=Whisper, Thresholds=(0.950, 0.950, 0.950)2026.04 | 67.3 | |
| PCA-based BaselineEmbedding Type=MFCC, Thresholds=(0.950, 0.950, 0.950)2026.04 | 65.6 | |
| Fixed-threshold BaselineEmbedding Type=Wav2Vec2, Thresholds=(0.900, 0.900, 0.900)2026.04 | 61.7 | |
| Fixed-threshold BaselineEmbedding Type=Whisper, Thresholds=(0.900, 0.900, 0.900)2026.04 | 61.3 | |
| Fixed-threshold BaselineEmbedding Type=MFCC, Thresholds=(0.900, 0.900, 0.900)2026.04 | 57.4 | |
| GPT-4oLLM Backbone=GPT-4o, Embedding Type=MFCC, Thresholds=(0.960, 0.950, 0.995)2026.04 | 39 | |
| Gemini-Pro-2.5LLM Backbone=Gemini-Pro-2.5, Embedding Type=MFCC, Thresholds=(0.970, 0.950, 0.998)2026.04 | 38.64 |