AI RESEARCH
DASH: Dynamic Audio-Driven Semantic Chunking for Efficient Omnimodal Token Compression
arXiv CS.AI
•
ArXi:2603.15685v1 Announce Type: cross Omnimodal large language models (OmniLLMs) jointly process audio and visual streams, but the resulting long multimodal token sequences make inference prohibitively expensive. Existing compression methods typically rely on fixed window partitioning and attention-based pruning, which overlook the piecewise semantic structure of audio-visual signals and become fragile under aggressive token reduction. We propose Dynamic Audio-driven Semantic cHunking (DASH), a.