Automatic Speech Translation (Ko2En) on Minds14
29.5BLEU ScoreWhisper
Evaluation results
| Method | Links | |
|---|---|---|
| WhisperModel Size=1.5B, #speech tokens/30 Sec.=15002026.01 | 29.5 | |
| FastSLMModel Size=4.8B, #speech tokens/30 Sec.=502026.01 | 28.9 | |
| Gemini-2.5-Flash#speech tokens/30 Sec.=9602026.01 | 26.3 | |
| Phi4-MMModel Size=5.8B, #speech tokens/30 Sec.=375, Fine-tuned (Korean dataset)=true2026.01 | 14.8 |