一作ACL 2026Findings· Findings of ACL
Temporal Contrastive Decoding: A Training-Free Method for Large Audio-Language Models
Yanda Li · Yuhan Liu · Zirui Song · Yunchao Wei · Martin Takáč · Salem Lahlou
摘要
Large audio-language models (LALMs) generalize across speech, sound, and music, but unified decoders can exhibit a temporal smoothing bias: transient acoustic cues may be underutilized in favor of temporally smooth context that is better supported by language priors, leading to less specific audi...