A contrastive-style adapter trained on LLM-generated positive and negative audio descriptions improves audio hallucination accuracy to 77.5 percent and audio question answering to 84.3 percent, without changing the frozen language model.
This is achieved by leveraging a backbone-LLM- synthesized dataset, which automatically generates audio- text pairs and contrastive data across general audio scenarios
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Teaching Audio-Aware Large Language Models What Does Not Hear: Mitigating Hallucinations through Synthesized Negative Samples
A contrastive-style adapter trained on LLM-generated positive and negative audio descriptions improves audio hallucination accuracy to 77.5 percent and audio question answering to 84.3 percent, without changing the frozen language model.