AI RESEARCH
The Cascade Equivalence Hypothesis: When Do Speech LLMs Behave Like ASR$\rightarrow$LLM Pipelines?
arXiv CS.AI
•
ArXi:2602.17598v2 Announce Type: replace-cross Speech LLMs are widely understood to be better than ASR$\rightarrow$LLM cascades since they have access to the audio directly, and not just the transcript. In this paper, we present an evaluation methodology and a mechanistic interpretation of the observed behavior of speech LLMs. First, we