ASR Hypothesis Selection on Kichwa (test)
12.4WEROracle
Evaluation Results
| Method | Links | |
|---|---|---|
| OracleDescription=Selection of best hypothesis by lowest WER against ground truth2026.01 | 12.4 | |
| Phi-4 TFTModel Category=Multimodal LLM, Fine-tuning Protocol=Target-language fine-tuning2026.01 | 16.6 | |
| LlamaModel Category=Text-based LLM, Variant=Llama3B2026.01 | 17 | |
| Phi-4 XFTModel Category=Multimodal LLM, Fine-tuning Protocol=Cross-lingual fine-tuning2026.01 | 17.1 | |
| Qwen3-OmniModel Category=Multimodal LLM2026.01 | 17.2 | |
| AcousticModel Category=Acoustic baseline2026.01 | 17.3 | |
| Phi-4Model Category=Multimodal LLM, Fine-tuning Protocol=None2026.01 | 17.4 | |
| Ngram-LMModel Category=Text-based LLM2026.01 | 17.7 | |
| Trans-LMModel Category=Text-based LLM2026.01 | 18.9 |