arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.18010cs.SD
收藏

通过多模态大语言模型扩展音频-文本检索

Re-purposing Multimodal Large Language Models for Audio-Text Retrieval

Visual Geometry Group, University of Oxford(牛津大学视觉几何组);Epidemic Sound;School of Artificial Intelligence, Shanghai Jiao Tong University(上海交通大学人工智能学院)

Jilan Xu, Carl Thomé, Danijela Horak, Weidi Xie, Andrew Zisserman

AI总结:

AuroLA通过多模态大语言模型实现音频-文本检索的扩展,利用可扩展的数据管道和混合NCE损失提升检索性能。

摘要:

音频-文本检索对于连接声学信号和自然语言至关重要。尽管对比双编码架构如CLAP已显示出前景,但它们本质上受限于小型编码器的容量。具体来说,文本编码器难以理解需要推理或世界知识的复杂查询。在本文中,我们提出了AuroLA,一种新颖的对比语言-音频预训练框架,该框架重新利用多模态大语言模型(MLLMs)作为统一的检索骨干。具体而言,我们做出了三个贡献:(i)我们构建了一个可扩展的数据管道,该管道从多个来源收集多样化的音频并生成多粒度的描述,从长描述到结构化标签,通过自动化标注;(ii)我们通过提示使MLLM总结音频/文本输入,并使用特殊标记的隐藏状态作为音频/文本嵌入。对于模型训练,我们设计了一种新颖的混合NCE损失,该损失采用多粒度监督和难负样本重新加权,以稳健地对齐音频与多样的文本监督;(iii)我们设计了一个基于MLLM的双向重排序模块,通过深度跨模态交互来优化检索候选。广泛的实验表明,AuroLA在性能上始终优于最先进的模型,包括最近的PE-AV,同时仅使用了PE-AV训练数据的大约1%。最后,我们观察到关于数据集大小和模型容量的明显扩展趋势,验证了MLLM作为音频-文本检索统一骨干的有效性。代码可在https://github.com/Jazzcharles/AuroLA上获取。

英文摘要:

Audio-text retrieval is crucial for bridging acoustic signals and natural language. While contrastive dual-encoder architectures like CLAP have shown promise, they are fundamentally limited by the capacity of small-scale encoders. Specifically, the text encoders struggle to understand complex queries that require reasoning or world knowledge. In this paper, we propose AuroLA, a novel contrastive language-audio pre-trained model that re-purposes Multimodal Large Language Models (MLLMs) as a unified backbone for audio-text retrieval. Specifically, we make the following contributions: (i) we construct a scalable data pipeline that curates diverse audio from multiple sources and generates multi-granular captions, ranging from long descriptions to structured tags, via automated annotation; (ii) we adapt an MLLM for retrieval by prompting it to summarise the audio/text input and using the hidden state of a special token as audio/text embeddings. (iii) extensive experiments demonstrate that AuroLA consistently outperforms state-of-the-art dual-encoder models, including the recent PE-AV. This validates the effectiveness of MLLM as a unified backbone for audio-text retrieval.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑