Baby_Cry_Detection_Database

Description

This dataset is a subset of the AudioSet, curated for Baby-cry detection task.
The curated dataset is divided into:

1. development set:
   The development dataset is further split into the TRAIN and VALIDATION sets.
2. evaluation set

The dataset has been categorized into two main classes: BabyCry and Other.
The Other class comprises of sound clips from the domestic household environment.
Note:
The recurrence of the events in the labels for each clip does not overlap.
The dataset contains an almost equal distribution of clips between BabyCry and Other classes.

The distribution for the dataset is as follows:

SET	development set		evaluation set
CLASS	TRAIN	VALIDATION	TEST
BabyCry	492	39	25
Other	480	39	40
TOTAL	972	78	65

TSV file format

The strong annotations are provided in a tab separated csv file under the following format:
[filename (string)][tab][event_label (string)][tab][onset (in seconds) (float)][tab][offset (in seconds) (float)][tab][start (in seconds) (float)][tab][name (string)][tab][original_label (string)]

filename: name of the audio file from where the 10-second clip was extracted t=start sec to t=start+10 sec, correspond to the clip boundaries within the full video.
event_label: updated class of the sound event as per our dataset distribution.
onset: onset time in seconds.
offset: offset time in seconds.
start: start of where the 10-second clip was extracted as t=start sec to t=start+10 sec.
name: name of the audio file.
original_label: original class of the sound event as present in AudioSet.

Dataset Download

requirements.txt is available to download the dependencies required to download the dataset.

After downloading the dependencies, you can download the dataset using the script: download_audioset_baby_cry_dataset.py.
It requires two arguements: workspace and data_type.

Usage:

Here are the commands to download train, validation, and test dataset respectively:
python download_audioset_baby_cry_dataset.py --workspace=$WORKSPACE --data_type=train
python download_audioset_baby_cry_dataset.py --workspace=$WORKSPACE --data_type=validation
python download_audioset_baby_cry_dataset.py --workspace=$WORKSPACE --data_type=test
(You can change workspace to the desired working folder.)
This will store the dowloaded files in the dataset folder created in the workspace under train, validation, and test respectively.

Summary

This codebase provides the dataset useful for baby cry studies in the domestic environment.

Citation

If this database is helpful, please feel free to cite the following paper:

Tanmay Khandelwal, Rohan Kumar Das, and Chng Eng Siong, “Is Your Baby Fine at Home? Baby Cry Sound Detection in Domestic Environments”, in Asia-Pacific Signal and Information Processing Association (APSIPA) Annual Summit and Conference (ASC), Chiang Mai, Thailand, pp. 275–280, 2022.
To access the paper
IEEEXplore || APSIPA ASC ||

BibTex reference
@INPROCEEDINGS{9980350, author={Khandelwal, Tanmay and Das, Rohan Kumar and Chng, Eng Siong}, booktitle={2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)}, title={Is Your Baby Fine at Home? Baby Cry Sound Detection in Domestic Environments}, year={2022}, volume={}, number={}, pages={275-280}, doi={10.23919/APSIPAASC55919.2022.9980350}}