---
tags:
- audio
- sound event detection
size_categories:
- 100K<n<1M
license: 
- cc-by-nc-sa-4.0
extra_gated_prompt: "The researcher has requested permission to use the ICSD dataset. In exchange for such permission, the researcher hereby agrees to the following terms and conditions:\n
1. Non-Commercial Use: The researcher shall use the dataset ONLY for non-commercial research and educational purposes.\n
2. No Distribution: The researcher shall not distribute, publish, or share the dataset, in whole or in part, to any third party without prior written consent from the dataset provider.\n
3. Modification and Derivative Works: The researcher may create modifications and derivative works of the dataset. However, any such works must be used solely for the purposes permitted by this agreement and must not be distributed or shared.\n
4. Acknowledgment of Source: Any publications or presentations resulting from the use of the dataset must acknowledge the ICSD dataset and its providers.
"
extra_gated_fields:
  Name: text
  Email: text
  Affiliation: text
  Position: text

  I agree to use this dataset for non-commercial use ONLY: checkbox
---
## ICSD: An Open-source Dataset for Infant Cry and Snoring Detection
<!-- ---
license: cc-by-nc-sa-4.0
--- -->

The ICSD dataset is a publicly available resource for the detection of infant cries and snoring sounds. It contains over 4 hours of audio data related to snoring sounds and infant crying sounds, as well as their corresponding annotations.

### Dataset Overview

> **Please note that our paper is currently under review. If you're interested in utilizing the dataset, please submit the necessary information on the top of this page to make a request. All the request will be reviewed and approved upon the acceptance of our paper.**

The ICSD dataset contains three kinds of data: real strongly labeled data, weakly labeled data and synthetic strongly labeled data. Real strongly labeled data, weakly labeled data come from Internet, and synthetic strongly labeled data is synthesized according to the collected foreground and background events. This variety ensures the dataset that can be used to train and evaluate models under different conditions, enhancing their robustness and generalizability. The entire dataset is divided into three subsets: train, validation and test sets for building the ICSD system. The table below provides the statistics for different types and subsets in the dataset.

<br />
<table border="1">
<thead>
<tr>
<th style="text-align:center"><strong>SET</strong></th>
<th colspan="2" style="text-align:center"><strong>train</strong></th>
<th colspan="2" style="text-align:center"><strong>validation</strong></th>
<th colspan="2" style="text-align:center"><strong>test</strong></th>
</tr>
</thead>
<tbody>
<tr>
<td style="text-align:center"><strong>TYPE</strong></td>
<td style="text-align:center" ><em>Infantcry</em></td>
<td style="text-align:center"><em>Snoring</em></td>
<td style="text-align:center" ><em>Infantcry</em></td>
<td style="text-align:center"><em>Snoring</em></td>
<td style="text-align:center" ><em>Infantcry</em></td>
<td style="text-align:center"><em>Snoring</em></td>
</tr>
<tr>
<td style="text-align:center"><strong>Weak label</strong></td>
<td style="text-align:center">1699</td>
<td style="text-align:center">1577</td>
<td style="text-align:center">189</td>
<td style="text-align:center">176</td>
<td style="text-align:center">None</td>
<td style="text-align:center">None</td>
</tr>
<tr>
<td style="text-align:center"><strong>Real strong label</strong></td>
<td style="text-align:center">338</td>
<td style="text-align:center">305</td>
<td style="text-align:center">43</td>
<td style="text-align:center">39</td>
<td style="text-align:center">43</td>
<td style="text-align:center">39</td>
</tr>
<tr>
<td style="text-align:center"><strong>Synthetic strong label</strong></td>
<td style="text-align:center">4000</td>
<td style="text-align:center">4000</td>
<td style="text-align:center">500</td>
<td style="text-align:center">500</td>
<td style="text-align:center">500</td>
<td style="text-align:center">500</td>
</tr>
</tbody>
</table>

### Dataset Usage
To use the ICSD dataset, you can download the audio files from the [provided URL list](https://huggingface.co/datasets/QingyuLiu1/ICSD) and the [code](https://github.com/QingyuLiu0521/ICSD/) we provided to run the baseline system.

The ICSD dataset will be structured as follows:
- Dataset ontology: the audio folder and its corresponding metadata folder
- Source materials: foreground and background events for generating synthetic strongly labeled data.

Please note that ICSD doesn't own the copyright of the audios; the copyright remains with the original owners of the audio.

### Reference
If you use the ICSD dataset, please cite the following papers:
```bibtex
@article{ICSD,
      title={ICSD: An Open-source Dataset for Infant Cry and Snoring Detection},
      author={Qingyu Liu, Longfei Song, Dongxing Xu, Yanhua Long},
      journal={arXiv},
      volume={abs/2408.10561}
      year={2024}
}
```