renumics/dcase23-task2-enriched
Dataset Card for the Enriched "DCASE 2023 Challenge Task 2 Dataset". Dataset Summary Data-centric AI principles have become increasingly important for real-world use cases. At Renumics we believe that classical benchmark datasets and competitions should be extended to reflect this development. This is why we are publishing benchmark datasets with application-specific enrichments (e.g. embeddings, baseline results, uncertainties, label error scores). We hope this… See the full description on the dataset page: https://huggingface.co/datasets/renumics/dcase23-task2-enriched.
6574
1---2license: cc-by-4.03task_categories:4- audio-classification5pretty_name: >-6 Enriched DCASE 2023 Challenge Task 2 Dataset7size_categories:8- 1K<n<10K9tags:10- anomaly detection 11- anomalous sound detection 12- acoustic condition monitoring 13- sound machine fault diagnosis 14- machine learning 15- unsupervised learning 16- acoustic scene classification 17- acoustic event detection 18- acoustic signal processing 19- audio domain shift 20- domain generalization21---22 23 24# Dataset Card for the Enriched "DCASE 2023 Challenge Task 2 Dataset".25 26## Table of contents27 28[//]: # (todo: create new)29- [Dataset Description](#dataset-description)30 - [Dataset Summary](#dataset-summary)31 - [Explore the data with Spotlight](#explore-the-data-with-spotlight)32- [Dataset Structure](#dataset-structure)33 - [Data Instances](#data-instances)34 - [Data Fields](#data-fields)35 - [Data Splits](#data-splits)36- [Dataset Creation](#dataset-creation)37 - [Curation Rationale](#curation-rationale)38 - [Source Data](#source-data)39 - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)40- [Considerations for Using the Data](#considerations-for-using-the-data)41 - [Social Impact of Dataset](#social-impact-of-dataset)42 - [Discussion of Biases](#discussion-of-biases)43 - [Other Known Limitations](#other-known-limitations)44- [Additional Information](#additional-information)45 - [Baseline system](#baseline-system)46 - [Dataset Curators](#dataset-curators)47 - [Licensing Information - Condition of use](#licensing-information---condition-of-use)48 - [Citation Information (original)](#citation-information-original)49 50## Dataset Description51 52- **Homepage:** [Renumics Homepage](https://renumics.com/)53- **Homepage** [DCASE23 Task 2 Challenge](https://dcase.community/challenge2023/task-first-shot-unsupervised-anomalous-sound-detection-for-machine-condition-monitoring#evaluation)54- **Homepage:** [HF Dataset Creator](https://syoy.github.io/)55- **Original Dataset Upload (Dev)** [ZENODO: DCASE 2023 Challenge Task 2 Development Dataset](https://zenodo.org/record/7687464#.Y_9VtdLMLmE)56- **Paper** [MIMII DG](https://arxiv.org/abs/2205.13879)57- **Paper** [ToyADMOS2](https://arxiv.org/abs/2106.02369)58- **Paper** [First-shot anomaly detection for machine condition monitoring: A domain generalization baseline](https://arxiv.org/pdf/2303.00455.pdf)59 60### Dataset Summary61 62[Data-centric AI](https://datacentricai.org) principles have become increasingly important for real-world use cases. At [Renumics](https://renumics.com/) we believe that classical benchmark datasets and competitions should be extended to reflect this development. 63 64This is why we are publishing benchmark datasets with application-specific enrichments (e.g. embeddings, baseline results, uncertainties, label error scores). We hope this helps the ML community in the following ways:651. Enable new researchers to quickly develop a profound understanding of the dataset.662. Popularize data-centric AI principles and tooling in the ML community.673. Encourage the sharing of meaningful qualitative insights in addition to traditional quantitative metrics.68 69This dataset is an enriched version of the [dataset](https://zenodo.org/record/7690148#.ZAXsSdLMLmE) provided in the context of the [anomalous sound detection task](https://dcase.community/challenge2023/task-first-shot-unsupervised-anomalous-sound-detection-for-machine-condition-monitoring) of the [DCASE2023 challenge](https://dcase.community/challenge2023/). The enrichment include an embedding generated by a pre-trained [Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer#transformers.ASTFeatureExtractor) and results of the official challenge [baseline implementation](https://github.com/nttcslab/dase2023_task2_baseline_ae).70 71### DCASE23 Task2 Dataset72 73Once a year, the [DCASE community](https://dcase.community/) publishes a [challenge](https://dcase.community/challenge2023/) with several tasks in the context of acoustic event detection and classification. [Task 2 of this challenge](https://dcase.community/challenge2023/task-first-shot-unsupervised-anomalous-sound-detection-for-machine-condition-monitoring) deals with anomalous sound detection for machine condition monitoring. The original dataset is based on the [MIMII DG](https://arxiv.org/abs/2205.13879) and the [ToyADMOS2](https://arxiv.org/abs/2106.02369) datasets. Please cite the papers by [Harada et al.](https://arxiv.org/abs/2106.02369) and [Dohi et al.](https://arxiv.org/abs/2205.13879) if you use this dataset and the paper by [Harada et al.](https://arxiv.org/pdf/2303.00455.pdf) if you use the baseline results.74 75### Explore Dataset76 7778 79The enrichments allow you to quickly gain insights into the dataset. The open source data curation tool Renumics Spotlight enables that with just a few lines of code:80 81Install datasets and Spotlight via [pip](https://packaging.python.org/en/latest/key_projects/#pip):82```python83!pip install renumics-spotlight datasets[audio]84```85> **_Notice:_** On Linux, non-Python dependency on libsndfile package must be installed manually. See [Datasets - Installation](https://huggingface.co/docs/datasets/installation#audio) for more information.86 87Load the dataset from huggingface in your notebook:88```python89import datasets90 91dataset = datasets.load_dataset("renumics/dcase23-task2-enriched", "dev", split="all", streaming=False)92```93 94Start exploring with a simple view that leverages embeddings to identify relevant data segments:95 96```python97from renumics import spotlight98 99df = dataset.to_pandas()100simple_layout = datasets.load_dataset_builder("renumics/dcase23-task2-enriched", "dev").config.get_layout(config="simple")101spotlight.show(df, dtype={'path': spotlight.Audio, "embeddings_ast-finetuned-audioset-10-10-0.4593": spotlight.Embedding}, layout=simple_layout)102```103You can use the UI to interactively configure the view on the data. Depending on the concrete taks (e.g. model comparison, debugging, outlier detection) you might want to leverage different enrichments and metadata.104 105In this example we focus on the valve class. We specifically look at normal data points that have high anomaly scores in both models. This is one example on how to find difficult example or edge cases:106 107 108```python109from renumics import spotlight110 111extended_layout = datasets.load_dataset_builder("renumics/dcase23-task2-enriched", "dev").config.get_layout(config="extended")112spotlight.show(df, dtype={'path': spotlight.Audio, "embeddings_ast-finetuned-audioset-10-10-0.4593": spotlight.Embedding}, layout=extended_layout)113```114 115116 117## Using custom model results and enrichments118 119When developing your custom model you want to use different kinds of information from you model (e.g. embedding, anomaly scores etc.) to gain further insights into the dataset and the model behvior.120 121Suppose you have your model's embeddings for each datapoint as a 2D-Numpy array called `embeddings` and your anomaly score as a 1D-Numpy array called `anomaly_scores`. Then you can add this information to the dataset:122```python123df['my_model_embedding'] = embeddings124df['anomaly_score'] = anomaly_scores125```126Depending on your concrete task you might want to use different enrichments. For a good overview on great open source tooling for uncertainty quantification, explainability and outlier detection, you can take a look at our [curated list for open source data-centric AI tooling](https://github.com/Renumics/awesome-open-data-centric-ai) on Github.127 128You can also save your view configuration in Spotlight in a JSON configuration file by clicking on the respective icon:129 130131 132For more information how to configure the Spotlight UI please refer to the [documentation](https://spotlight.renumics.com).133 134## Dataset Structure135 136### Data Instances137 138For each instance, there is a Audio for the audio, a string for the path, an integer for the section, a string for the d1p (parameter), a string for the d1v (value),139a ClassLabel for the label and a ClassLabel for the class.140 141```python142{'audio': {'array': array([ 0. , 0.00024414, -0.00024414, ..., -0.00024414,143 0. , 0. ], dtype=float32),144 'path': 'train/fan_section_01_source_train_normal_0592_f-n_A.wav',145 'sampling_rate': 16000146 }147 'path': 'train/fan_section_01_source_train_normal_0592_f-n_A.wav'148 'section': 1149 'd1p': 'f-n'150 'd1v': 'A'151 'd2p': 'nan'152 'd2v': 'nan'153 'd3p': 'nan'154 'd3v': 'nan'155 'domain': 0 (source)156 'label': 0 (normal)157 'class': 1 (fan)158 'dev_train_lof_anomaly': 0159 'dev_train_lof_anomaly_score': 1.241023160 'add_train_lof_anomaly': 1161 'add_train_lof_anomaly_score': 1.806289162 'ast-finetuned-audioset-10-10-0.4593-embeddings': [0.8152204155921936,163 1.5862374305725098, ..., 164 1.7154160737991333]165}166```167 168The length of each audio file is 10 seconds.169 170### Data Fields171 172- `audio`: an `datasets.Audio` 173- `path`: a string representing the path of the audio file inside the _tar.gz._-archive.174- `section`: an integer representing the section, see [Definition](#Description)175- `d*p`: a string representing the name of the d*-parameter176- `d*v`: a string representing the value of the corresponding d*-parameter177- `domain`: an integer whose value may be either _0_, indicating that the audio sample is from the _source_ domain, _1_, indicating that the audio sample is from the _target_.178- `class`: an integer as class label.179- `label`: an integer whose value may be either _0_, indicating that the audio sample is _normal_, _1_, indicating that the audio sample contains an _anomaly_.180- '[X]_lof_anomaly': an integer as anomaly indicator. The anomaly prediction is computed with the [Local Outlier Factor](https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.LocalOutlierFactor.html) algorithm based on the "[X]"-dataset.181- '[X]_lof_anomaly_score': a float as anomaly score. The anomaly score is computed with the [Local Outlier Factor](https://scikit-learn.org/stable/modules/generated/sklearn.neighbors.LocalOutlierFactor.html) algorithm based on the "[X]"-dataset.182- `embeddings_ast-finetuned-audioset-10-10-0.4593`: an `datasets.Sequence(Value("float32"), shape=(1, 768))` representing audio embeddings that are generated with an [Audio Spectrogram Transformer](https://huggingface.co/docs/transformers/model_doc/audio-spectrogram-transformer#transformers.ASTFeatureExtractor).183 184### Data Splits185 186The development dataset has 2 splits: _train_ and _test_.187 188| Dataset Split | Number of Instances in Split | Source Domain / Target Domain Samples |189| ------------- |------------------------------|---------------------------------------|190| Train | 7000 | 6930 / 70 |191| Test | 1400 | 700 / 700 |192 193The additional training dataset has 1 split: _train_.194 195| Dataset Split | Number of Instances in Split | Source Domain / Target Domain Samples |196| ------------- |------------------------------|---------------------------------------|197| Train | 7000 | 6930 / 70 |198 199The evaluation dataset has 1 split: _test_.200 201| Dataset Split | Number of Instances in Split | Source Domain / Target Domain Samples |202|---------------|------------------------------|---------------------------------------|203| Test | 1400 | ? |204 205 206## Dataset Creation207 208The following information is copied from the original [dataset upload on zenodo.org](https://zenodo.org/record/7690148#.ZAXsSdLMLmE)209 210### Curation Rationale211 212This dataset is the "development dataset" for the [DCASE 2023 Challenge Task 2 "First-Shot Unsupervised Anomalous Sound Detection for Machine Condition Monitoring"](https://dcase.community/challenge2023/task-unsupervised-anomalous-sound-detection-for-machine-condition-monitoring).213 214The data consists of the normal/anomalous operating sounds of seven types of real/toy machines. Each recording is a single-channel 10-second audio that includes both a machine's operating sound and environmental noise. The following seven types of real/toy machines are used in this task:215 216- ToyCar217- ToyTrain218- Fan219- Gearbox220- Bearing221- Slide rail222- Valve223 224The "additional training data" and "evaluation data" datasets contain the following classes:225 226- bandsaw227- grinder228- shaker229- ToyDrone 230- ToyNscale231- ToyTank232- Vacuum233 234### Source Data235 236#### Definition237 238We first define key terms in this task: "machine type," "section," "source domain," "target domain," and "attributes.".239 240- "Machine type" indicates the type of machine, which in the development dataset is one of seven: fan, gearbox, bearing, slide rail, valve, ToyCar, and ToyTrain.241- A section is defined as a subset of the dataset for calculating performance metrics.242- The source domain is the domain under which most of the training data and some of the test data were recorded, and the target domain is a different set of domains under which some of the training data and some of the test data were recorded. There are differences between the source and target domains in terms of operating speed, machine load, viscosity, heating temperature, type of environmental noise, signal-to-noise ratio, etc.243- Attributes are parameters that define states of machines or types of noise.244 245#### Description246 247This dataset consists of seven machine types. For each machine type, one section is provided, and the section is a complete set of training and test data. For each section, this dataset provides (i) 990 clips of normal sounds in the source domain for training, (ii) ten clips of normal sounds in the target domain for training, and (iii) 100 clips each of normal and anomalous sounds for the test. The source/target domain of each sample is provided. Additionally, the attributes of each sample in the training and test data are provided in the file names and attribute csv files.248 249#### Recording procedure250 251Normal/anomalous operating sounds of machines and its related equipment are recorded. Anomalous sounds were collected by deliberately damaging target machines. For simplifying the task, we use only the first channel of multi-channel recordings; all recordings are regarded as single-channel recordings of a fixed microphone. We mixed a target machine sound with environmental noise, and only noisy recordings are provided as training/test data. The environmental noise samples were recorded in several real factory environments. We will publish papers on the dataset to explain the details of the recording procedure by the submission deadline.252 253### Supported Tasks and Leaderboards254 255Anomalous sound detection (ASD) is the task of identifying whether the sound emitted from a target machine is normal or anomalous. Automatic detection of mechanical failure is an essential technology in the fourth industrial revolution, which involves artificial-intelligence-based factory automation. Prompt detection of machine anomalies by observing sounds is useful for monitoring the condition of machines. 256 257This task is the follow-up from DCASE 2020 Task 2 to DCASE 2022 Task 2. The task this year is to develop an ASD system that meets the following four requirements.258 259**1. Train a model using only normal sound (unsupervised learning scenario)**260 261Because anomalies rarely occur and are highly diverse in real-world factories, it can be difficult to collect exhaustive patterns of anomalous sounds. Therefore, the system must detect unknown types of anomalous sounds that are not provided in the training data. This is the same requirement as in the previous tasks.262 263**2. Detect anomalies regardless of domain shifts (domain generalization task)** 264 265In real-world cases, the operational states of a machine or the environmental noise can change to cause domain shifts. Domain-generalization techniques can be useful for handling domain shifts that occur frequently or are hard-to-notice. In this task, the system is required to use domain-generalization techniques for handling these domain shifts. This requirement is the same as in DCASE 2022 Task 2.266 267**3. Train a model for a completely new machine type**268 269For a completely new machine type, hyperparameters of the trained model cannot be tuned. Therefore, the system should have the ability to train models without additional hyperparameter tuning.270 271**4. Train a model using only one machine from its machine type**272 273While sounds from multiple machines of the same machine type can be used to enhance detection performance, it is often the case that sound data from only one machine are available for a machine type. In such a case, the system should be able to train models using only one machine from a machine type.274 275## Considerations for Using the Data276 277### Social Impact of Dataset278 279[More Information Needed]280 281### Discussion of Biases282 283[More Information Needed]284 285### Other Known Limitations286 287[More Information Needed]288 289## Additional Information290 291### Baseline system 292 293The baseline system is available on the Github repository [dcase2023_task2_baseline_ae](https://github.com/nttcslab/dase2023_task2_baseline_ae).The baseline systems provide a simple entry-level approach that gives a reasonable performance in the dataset of Task 2. They are good starting points, especially for entry-level researchers who want to get familiar with the anomalous-sound-detection task.294 295### Dataset Curators296 297[//]: # (todo)298 299[More Information Needed]300 301### Licensing Information - Condition of use302 303This is a feature/embeddings-enriched version of the "DCASE 2023 Challenge Task 2 Development Dataset".304The [original dataset](https://dcase.community/challenge2023/task-first-shot-unsupervised-anomalous-sound-detection-for-machine-condition-monitoring#audio-datasets) was created jointly by **Hitachi, Ltd.** and **NTT Corporation** and is available under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.305 306 307### Citation Information (original)308 309If you use this dataset, please cite all the following papers. We will publish a paper on DCASE 2023 Task 2, so pleasure make sure to cite the paper, too.310 311- Kota Dohi, Tomoya Nishida, Harsh Purohit, Ryo Tanabe, Takashi Endo, Masaaki Yamamoto, Yuki Nikaido, and Yohei Kawaguchi. MIMII DG: sound dataset for malfunctioning industrial machine investigation and inspection for domain generalization task. In arXiv e-prints: 2205.13879, 2022. [[URL](https://arxiv.org/abs/2205.13879)]312- Noboru Harada, Daisuke Niizumi, Daiki Takeuchi, Yasunori Ohishi, Masahiro Yasuda, and Shoichiro Saito. ToyADMOS2: another dataset of miniature-machine operating sounds for anomalous sound detection under domain shift conditions. In Proceedings of the 6th Detection and Classification of Acoustic Scenes and Events 2021 Workshop (DCASE2021), 1–5. Barcelona, Spain, November 2021. [[URL](https://dcase.community/documents/workshop2021/proceedings/DCASE2021Workshop_Harada_6.pdf)]313- Noboru Harada, Daisuke Niizumi, Daiki Takeuchi, Yasunori Ohishi, and Masahiro Yasuda. First-shot anomaly detection for machine condition monitoring: a domain generalization baseline. In arXiv e-prints: 2303.00455, 2023. [[URL](https://arxiv.org/abs/2303.00455.pdf)]314 315```316@dataset{kota_dohi_2023_7882613,317 author = {Kota Dohi and318 Keisuke Imoto and319 Noboru Harada and320 Daisuke Niizumi and321 Yuma Koizumi and322 Tomoya Nishida and323 Harsh Purohit and324 Takashi Endo and325 Yohei Kawaguchi},326 title = {DCASE 2023 Challenge Task 2 Development Dataset},327 month = mar,328 year = 2023,329 publisher = {Zenodo},330 version = {3.0},331 doi = {10.5281/zenodo.7882613},332 url = {https://doi.org/10.5281/zenodo.7882613}333}334```