[论文] AVGen-Bench: A Task-Driven Benchmark for Multi-Granular Evaluation of ...

## 论文概要 **研究领域**: NLP **作者**: Ziwei Zhou, Zeyuan Lai, Rui Wang **发布时间**: 2025-04-10 **arXiv**: [2504.07857](https://arxiv.org/abs/2504.07857) ## 中文摘要 文本到音视频(T2AV)生成正迅速成为媒体创作的核心接口，但其评估仍然碎片化。现有基准测试主要孤立地评估音频和视频，或依赖粗略的嵌入相似性，无法捕捉真实提示词所需的细粒度联合正确性。我们推出了AVGen-Bench，一个用于T2AV生成的任务驱动基准测试，包含跨11个真实世界类别的高质量提示词。为支持全面评估，我们提出了一个多粒度评估框架，结合轻量级专家模型与多模态大语言模型(MLLM)，实现从感知质量到细粒度语义可控性的评估。我们的评估揭示了强音视频美学与弱语义可靠性之间的显著差距，包括在文本渲染、语音连贯性、物理推理方面的持续失败，以及音乐音高控制的普遍失效。 ## 原文摘要 Text-to-Audio-Video (T2AV) generation is rapidly becoming a core interface for media creation, yet its evaluation remains fragmented. Existing benchmarks largely assess audio and video in isolation or rely on coarse embedding similarity, failing to capture the fine-grained joint correctness required by realistic prompts. We introduce AVGen-Bench, a task-driven benchmark for T2AV generation featuring high-quality prompts across 11 real-world categories. To support comprehensive assessment, we propose a multi-granular evaluation framework that combines lightweight specialist models with Multimodal Large Language Models (MLLMs), enabling evaluation from perceptual quality to fine-grained semantic controllability. Our evaluation reveals a pronounced gap between strong audio-visual aesthetics a... --- *自动采集于 2026-04-12* #论文 #arXiv #NLP #小凯