Speech Recognition on Vaani 48 languages (test)
27.4WERSigLIP2-large
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SigLIP2-largeinput_mode=multi-image, alignment=Audio-Image Alignment2026.06 | 27.4 | 0.0069 | 2.47 | 0.0062 | |
| Qwen3-VLinput_mode=multi-image, alignment=Audio-Image Alignment2026.06 | 27.68 | 0.0041 | 1.47 | 0.0033 | |
| SigLIP2-baseinput_mode=single-image, alignment=Audio-Image Alignment2026.06 | 27.71 | 0.0038 | 1.35 | 0.003 | |
| Baselinealignment=no alignment2026.06 | 28.09 | — | — | — |