Speech Recognition on German
8.53WERWhisper
Evaluation results
| Method | Links | |
|---|---|---|
| Whisper2026.06 | 8.53 | |
| BLoRA# utterances=246k2026.06 | 9.29 | |
| BLoRA# utterances=20k2026.06 | 9.31 | |
| BLoRA# utterances=10k2026.06 | 9.77 | |
| BLoRA# utterances=1k2026.06 | 11.59 | |
| LLM-P2G + SKMFramework=LLM-P2G, Training Strategy=Sampling-K Marginalized (SKM)2025.12 | 12.94 | |
| LoRA# utterances=246k2026.06 | 13 | |
| LLM-P2G + randomized TKMFramework=LLM-P2G, Training Strategy=Randomized TKM2025.12 | 13.03 | |
| Whistle + pooling-adapter + mT5Model Type=SpeechLLM, Acoustic Model=Whistle, Adapter=Pooling Adapter, LLM=mT52025.12 | 13.35 | |
| LLM-P2G + DANPFramework=LLM-P2G, Training Strategy=Data Augmentation with Noisy Phonemes (DANP)2025.12 | 13.63 | |
| LoRA# utterances=20k2026.06 | 17.69 | |
| LoRA# utterances=10k2026.06 | 20.8 | |
| 1word# utterances=10k2026.06 | 22.54 | |
| 0.2# utterances=10k2026.06 | 26.84 | |
| 3word# utterances=10k2026.06 | 30.78 | |
| LoRA# utterances=1k2026.06 | 44.15 |