REVE: Efficient Hallucination Correction for Large Audio-Language Models via Reused Encoder States

T1Covered by 1 source · first reported by arXiv at 24 Sept, 04:00 UTC

arXiv:2609.26028v1 Announce Type: cross Abstract: Large audio-language models may mention acoustic events that are absent from the input. A separate audio event detector can verify these mentions, but doing so requires a second audio encoder and a separate forward pass. We propose Reused Encoder…

science

Sources

• T1REVE: Efficient Hallucination Correction for Large Audio-Language Models via Reused Encoder StatesarXiv24 Sept, 04:00