ToyADMOS

ToyADMOS 数据集是一个机器操作声音数据集，包含大约 540 小时的正常机器操作声音和超过 12,000 个异常声音样本，由 Yuma Koizumi 和 NTT 媒体情报实验室的成员以 48kHz 的采样率收集的四个麦克风。 ToyADMOS 数据集专为机器操作声音 (ADMOS) 研究中的异常检测而设计。我们通过故意损坏它们的组件来收集微型机器的正常和异常操作声音。它是为ADMOS的三个任务而设计的：产品检测（玩具车）、固定机器（玩具输送机）的故障诊断和移动机器（玩具火车）的故障诊断。有关详细信息，请参阅论文 [1]。如果您在工作中使用 ToyADMOS 数据集，请在介绍它的地方引用本文。The ToyADMOS Dataset is a machine operating sound dataset containing approximately 540 hours of normal machine operating sounds and over 12,000 anomalous sound samples. It was collected by Yuma Koizumi and members of the NTT Media Intelligence Laboratories using four microphones at a sampling rate of 48 kHz. The ToyADMOS Dataset is specifically designed for anomaly detection research in machine operating sound (ADMOS) scenarios. Normal and anomalous operating sounds of miniature machines were acquired by intentionally damaging their components. This dataset is developed for three ADMOS tasks: product inspection (toy car), fault diagnosis of stationary machinery (toy conveyor belt), and fault diagnosis of mobile machinery (toy train). For detailed information, please refer to the paper [1]. If you use the ToyADMOS Dataset in your work, please cite this paper when introducing it.

搜集汇总

背景与挑战

背景概述

ToyADMOS是一个专注于机器操作声音异常检测的数据集，包含约540小时的正常声音和超过12,000个异常样本，采样率为48kHz，适用于产品检测、固定和移动机器的故障诊断任务。该数据集设计用于无监督异常检测和开放集学习，由立命馆大学和NTT于2019年发布，旨在支持音频异常检测研究。