Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
184 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

ToyADMOS: A Dataset of Miniature-Machine Operating Sounds for Anomalous Sound Detection (1908.03299v1)

Published 9 Aug 2019 in eess.AS, cs.LG, cs.SD, and stat.ML

Abstract: This paper introduces a new dataset called "ToyADMOS" designed for anomaly detection in machine operating sounds (ADMOS). To the best our knowledge, no large-scale datasets are available for ADMOS, although large-scale datasets have contributed to recent advancements in acoustic signal processing. This is because anomalous sound data are difficult to collect. To build a large-scale dataset for ADMOS, we collected anomalous operating sounds of miniature machines (toys) by deliberately damaging them. The released dataset consists of three sub-datasets for machine-condition inspection, fault diagnosis of machines with geometrically fixed tasks, and fault diagnosis of machines with moving tasks. Each sub-dataset includes over 180 hours of normal machine-operating sounds and over 4,000 samples of anomalous sounds collected with four microphones at a 48-kHz sampling rate. The dataset is freely available for download at https://github.com/YumaKoizumi/ToyADMOS-dataset

Citations (164)

Summary

  • The paper introduces the ToyADMOS dataset for advancing anomaly detection in miniature machine sounds by providing extensive normal and anomalous recordings.
  • The dataset comprises three sub-datasets—toy cars, conveyors, and trains—captured with four-channel 48-kHz audio under controlled conditions.
  • The collection enables exploration of domain adaptation, noise reduction, and few-shot learning, offering practical insights for ADMOS research.

A Study on ToyADMOS: A Dataset for Anomalous Sound Detection in Miniature Machines

In the domain of acoustic anomaly detection, the ToyADMOS dataset occupies a significant position as it addresses a critical gap. Acknowledging the scarcity of large-scale anomaly detection in machine operating sounds (ADMOS) datasets, this paper introduces "ToyADMOS," an extensive dataset designed to enhance the training and evaluation processes of ADMOS systems. The dataset comprises sound recordings from miniature machines—specifically, toys—experiencing both normal and deliberately induced anomalous operational conditions.

Dataset Composition and Characteristics

The ToyADMOS dataset stands out due to its scale and controlled creation methodology. It includes three sub-datasets: one for product inspection using toy cars, another for fault diagnosis of fixed machines using toy conveyors, and a third for fault diagnosis of moving machines using toy trains. Over 180 hours of normal operational sounds and more than 4,000 samples of anomalous sounds are compiled using four-channel 48-kHz recordings.

Each sub-dataset is meticulously configured:

  • Toy Car Sub-dataset: Models product inspection tasks, resulting in data from toy cars on inspection devices.
  • Toy Conveyor Sub-dataset: Focuses on fault diagnosis for static machines using toy conveyor systems.
  • Toy Train Sub-dataset: Targets fault diagnosis in moving machines with model trains on railway tracks.

The controlled settings where the data is collected include individually recordable machine-operating sounds and corresponding environmental noise, accommodating different noise levels.

Technical Advancements and Usability

The data collected serve various purposes in ADMOS research, from unsupervised anomaly detection to more advanced analyses like domain adaptation, noise reduction, and few-shot learning. Here are the implications:

  • Domain Adaptation: Different machines of the same class but varied structures enable the testing of systems for domain adaptation and individual machine variations.
  • Noise Reduction and Augmentation: The data's multi-channel nature facilitates the experimentation with noise reduction strategies.
  • Few-Shot Learning: By providing multiple examples of the same anomaly, the dataset aids the development of models that can generalize and adapt to sparse data environments.

Impact on ADMOS Research and Future Prospects

The ToyADMOS dataset is poised to be a catalyst for advancements in ADMOS, allowing for a deeper understanding of sound-based anomaly detection in machine sounds. The dataset's free availability ensures broad access to valuable resources, fostering innovation and potentially leading to significant improvements in practical applications such as condition monitoring and preventive maintenance.

The dataset’s controlled environment may not fully capture the complexities of real-world machine sounds. However, it provides a significant approximation, essential for the development and preliminary testing of anomaly detection algorithms. Future work could involve integrating this dataset with real-world data to create hybrid models that leverage controlled conditions for foundational learning and real-world data for model robustification.

In conclusion, the ToyADMOS dataset marks a substantial stride in arming researchers with the tools necessary to propel anomaly detection in acoustic domains forward. It provides a foundation not just for benchmarking but for crafting novel methodologies that could reshape how industries monitor and maintain machine operations.

Github Logo Streamline Icon: https://streamlinehq.com