IDMT-ISA-Metal-Balls Dataset

The IDMT_ISA_METAL_BALLS dataset contains sound files of three differently coated types of metal balls running down a metal slide (K1, K2, K3). The different surfaces cause a hearable difference of the rolling sounds, which we consider as indicator regarding the condition of the coating. The dataset was created using a low-cost microphone as sensor. Industrial background noise was played via external loudspeakers during the recording process for simulating a realistic acoustic environment. We provide both an unbalanced split, which resulted from the original recording process, with a total size of 2267 recordings as well as a balanced split for reproducibility of experiments. Two additional test sets with more than 250 recordings each are provided where general settings like the angle of the slide and the position of microphone have been altered for testing the robustness of the system to changes in the setup. File duration: 18.87 minutes # K1: 987 # K2: 1198 # K3: 647 # Total WAV Files: 2832 Sampling rate: 44.1KHz Resolution: 32-bit Mono audioIDMT_ISA_METAL_BALLS数据集收录了三类经不同涂层处理的金属球沿金属滑道（K1、K2、K3）滚落的音频文件。不同的涂层表面会产生可听辨的滚动声学差异，我们将该差异作为表征涂层状态的指示性指标。该数据集采用低成本麦克风作为传感器采集构建，录制过程中通过外置扬声器播放工业背景噪声，以模拟真实的工业声学环境。本次公开的数据集包含两种划分方案：其一为原始录制流程生成的非平衡划分，总计包含2267条录音；其二为保障实验可复现性而构建的平衡划分。此外还提供了两组额外测试集，每组均包含超过250条录音，测试集中调整了滑道角度、麦克风位置等通用采集参数，用于测试系统对采集设置变化的鲁棒性。音频文件总时长为18.87分钟，其中K1类录音987条、K2类1198条、K3类647条，总WAV文件数为2832条。采样率为44.1kHz，分辨率为32位单声道音频。