AI RESEARCH

FastSLM: Hierarchical Temporal Abstraction for Efficient Long-Form Speech Adaptation

arXiv CS.AI

ArXi:2601.06199v3 Announce Type: replace-cross Scaling Multimodal Large Language Models (MLLMs) to long-form speech is bottlenecked by the explosive growth of input tokens. Unlike images or videos, audio lacks overlapping information, making extreme 1-token compression highly susceptible to the loss of fine-grained acoustic cues. To overcome this, we propose FastSLM, a token-efficient architecture featuring the Hierarchical Temporal Abstractor