1 paper · 1 filter
Chih-Kai Yang, Neo Ho, Yi-Jyun Lee +1
Understanding the internal mechanisms of large audio-language models (LALMs) is crucial for interpreting their behavior and improving performance. This work presents the first in-d…