arXiv cs.CL· Eunsik Shin·· 13 小时前AI 评分31
冻结的大型音频语言模型能验证音乐时刻却难以定位：提示词低估了时间信息
Heard But Not Said: Prompting Understates Temporal Information of Musical Moments in Frozen Large Audio-Language Models
AI 导读
研究用 Slakh2100 分轨激活构建配对查询探针发现，冻结的大型音频语言模型（LALM）在无上下文时状态查询得分高于基线，而变化查询在多数短片段模型上随前文上下文改善，30s padded 编码器则无类似恢复。提示词与注意力两种方式几乎都暴露不出这一能力，说明提示失败低估了模型可验证的时间信息，但并未低估其定位能力，差距大小由查询类型和编码器决定。
正文 · 原文
Abstract:Large audio-language models (LALMs) understand music, yet struggle to locate a described moment in time. We ask whether this reflects missing temporal information or a bottleneck in how the time span is extracted. Using a paired-query probe built from Slakh2100 stem activations, we attach a state query (e.g., ``only piano and bass are playing'') and a change query (e.g., ``the guitar drops out, leaving piano and bass'') to the same span and vary how much preceding audio the model is shown. Across frozen LALMs, three ways of extracting the span ask different questions: full-track prompting asking whether the model can say when; windowed yes/no scoring asking whether it verifies the moment when handed the segment, and whether it needs what came before; attention asking whether it can find the moment on its own. State queries score above baseline without context, while change queries improve with preceding context in most short-segment models; 30s padded encoders show no comparable recovery. Prompting and attention expose almost none of this. Prompting failures thus understate what LALMs can verify, though not what they can locate, by an amount set by query type and encoder.
| Subjects: | Sound (cs.SD); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.29256 [cs.SD] |
| (or arXiv:2610.29256v1 [cs.SD] for this version) | |
| https://doi.org/10.48550/arXiv.2610.29256 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Eunsik Shin [view email]
[v1]
Wed, 21 Oct 2026 06:06:11 UTC (50 KB)
来源：arXiv cs.CL · arxiv.org