Intent Recognition on VoiceStick (Full Set)
92AccuracyCamemBERT (Large)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CamemBERT (Large)Input=Manual Text (Oracle)2026.06 | 92 | 22 | |
| Average⋆Input=Manual Text (Oracle)2026.06 | 86 | 16 | |
| XLSR-53-FR (Large)Input=Raw Audio (Proposed End-to-End), Distillation=Yes2026.06 | 82 | 7 | |
| XLSR-53-FR (Large)Input=Raw Audio, Distillation=No2026.06 | 65 | 7 | |
| WavLM (Large)Input=Raw Audio (Proposed End-to-End), Distillation=Yes2026.06 | 64 | 14 | |
| Average⋆Input=Raw Audio (Proposed End-to-End), Distillation=Yes2026.06 | 64 | 11 | |
| CamemBERT (Large)Input=Audio → Whisper → Text (Cascade)2026.06 | 59 | 202 | |
| WavLM (Large)Input=Raw Audio, Distillation=No2026.06 | 59 | 14 | |
| Average⋆Input=Audio → Whisper → Text (Cascade)2026.06 | 52 | 196 | |
| Average⋆Input=Raw Audio, Distillation=No2026.06 | 52 | 11 | |
| XLSR-53 (Large)†Input=Raw Audio (Proposed End-to-End), Distillation=Yes2026.06 | 50 | 10 |