Temporal Grounding (Word Alignment) on Librispeech
7.7Accuracy (20ms tolerance)Gemini 2.5 Flash
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Gemini 2.5 FlashInference Mode=Zero-shot2026.02 | 7.7 | 13.8 | 0.28 | |
| Voxtral 3BInference Mode=Zero-shot2026.02 | 0.8 | 1.1 | 2.52 | |
| Qwen2.5 7BInference Mode=Zero-shot2026.02 | 0.7 | 1.1 | 2.17 | |
| Audio Flamingo 3Inference Mode=Zero-shot2026.02 | 0.6 | 1.4 | 1.82 | |
| GPT-4o AudioInference Mode=Zero-shot2026.02 | 0.3 | 0.5 | 2.42 | |
| Voxtral 24BInference Mode=Zero-shot2026.02 | 0.2 | 0.5 | 2.12 | |
| Qwen2.5 3BInference Mode=Zero-shot2026.02 | 0.1 | 0.2 | 2.59 |