遇见数据集

LiuZH-19/starbech_tmp

收藏
Hugging Face2025-10-23 更新2025-10-25 收录
官方服务:

资源简介:

STAR-Bench是一个衡量音频4D智能的数据集,它通过基础声音感知设置和整体时空推理设置来评估模型对声音在时间和3D空间中的动态推理能力。数据集包括对连续和离散过程的段重新排序以及静态定位、多源关系和动态轨迹的空间任务。该数据集的构成经过专家验证和人类测试校准。

STAR-Bench is a benchmark for measuring audio 4D intelligence, which evaluates the models ability to reason about sound dynamics in time and 3D space through foundational acoustic perception settings and holistic spatio-temporal reasoning settings. The dataset includes segment reordering for continuous and discrete processes, as well as spatial tasks such as static localization, multi-source relationships, and dynamic trajectories. The construction of this dataset has been validated by experts and calibrated through human testing.

提供机构:
LiuZH-19
二维码
社区交流群
二维码
科研交流群
商业服务