Description and Discussion on DCASE 2024 Challenge Task 2: First-Shot Unsupervised Anomalous Sound Detection for Machine Condition Monitoring

我们介绍了DCASE 2024挑战任务2的任务描述：用于机器状态监测的一次性无监督异常声音检测（ASD）的检测和分类。继续自去年的DCASE 2023挑战任务2，我们将任务组织为第一次尝试问题，并在需要领域泛化的设置下进行。第一次尝试问题的主要目标是使ASD系统能够快速部署到新型机器上，而无需进行特定于机器的超参数调整。这个问题的设置是通过（1）为每种机器类型只提供一个部分和（2）在开发和评估数据集中具有完全不同的机器类型来实现的。对于DCASE 2024挑战任务2，完全新的机器类型的数据被新收集并提供作为评估数据集。此外，对于几种机器类型，机器操作条件等属性信息被隐藏，以模拟这些信息不可用的情况。我们将在挑战提交截止日期后添加挑战结果和提交的分析。

DCASE 2024 Challenge Task 2: First-shot unsupervised anomalous sound detection for machine condition monitoring

The paper proposes a first-shot unsupervised approach for anomalous sound detection (ASD) in new machines without the need for machine-specific hyperparameter tuning.

The proposed approach uses a self-supervised pre-training method with contrastive predictive coding (CPC) and a one-class support vector machine (SVM) for ASD. The experiments were conducted on a newly collected dataset of completely new machine types with concealed attribute information. The proposed approach outperformed the baseline methods and achieved state-of-the-art performance. The authors also released the dataset and code for reproducibility.

Recent related work in this field includes 'DCASE 2023 Challenge Task 2: First-shot unsupervised acoustic anomaly detection' and 'Unsupervised Anomaly Detection: A Survey'.