Localizing and Editing Knowledge in Large Audio-Language Models
This work addresses the issue of updating factual knowledge in speech AI systems for users relying on LALMs for information access, representing an incremental advancement by adapting existing model editing methods to handle continuous speech representations.
The paper tackles the problem of incorrect factual knowledge in Large Audio-Language Models (LALMs) by developing a speech-driven locate-then-edit framework, which localizes knowledge across audio and text modules and enables more effective updates through audio editing than text editing or fine-tuning.
Large Audio-Language Models (LALMs) have shown strong performance in speech understanding, making speech a natural interface for accessing factual information. Yet they are trained on static corpora and may encode incorrect facts. Existing model editing methods localize and update facts in text-only LLMs, but do not account for continuous speech representations, or where knowledge is stored across acoustic or language modules, or their cross-modal module. We construct the first audio benchmark for knowledge localization and editing in LALMs and propose a speech-driven locate-then-edit framework. First, we use speech-aware causal tracing to localize layers and modules that support factual retrieval and then apply editing at identified sites. Experiments show that factual knowledge is jointly encoded in audio and text modules, and that audio editing yields more effective updates than text editing or fine-tuning, enabling fine-grained knowledge control in speech AI systems.