Speech-to-Text Question-Answering on LlamaQ, TriviaQA, WebQ, and OBQA Average
57.6AccuracyDIFFUSPEECH
Evaluation Results
| Method | Links | |
|---|---|---|
| DIFFUSPEECHType=Diff., S→S capability=true2026.01 | 57.6 | |
| MinMoType=AR, S→S capability=true2026.01 | 54 | |
| Qwen2-AudioType=AR, S→S capability=false2026.01 | 46.2 | |
| MoshiType=AR, S→S capability=true2026.01 | 46 | |
| Phi-4-MultimodalType=AR, S→S capability=false2026.01 | 43.9 | |
| Llama-Omni2Type=AR, S→S capability=true2026.01 | 43.5 | |
| DiFFAType=Diff., S→S capability=false2026.01 | 43.3 | |
| SpiritLMType=AR, S→S capability=true2026.01 | 20.2 | |
| SpeechGPTType=AR, S→S capability=true2026.01 | 13.8 |