Datasets:

chimechallenge
/

chime3-background-noise

Modalities:

Audio

Formats:

Size:

Tags:

Libraries:

License:

Dataset card Data Studio Files Files and versions

xet

Community

Dataset Viewer

Auto-converted to Parquet Duplicate

Split (1)

train · 102 rows

Search is not available for this dataset

audio audioduration (s) 1.01k 2.87k

End of preview. Expand in Data Studio

CHiME-3 Multichannel Background Noise Recordings

This repository mirrors the dataset archived on Zenodo at https://doi.org/10.5281/zenodo.23571515. Please cite that Zenodo record when using these data. No separate DOI is minted for this Hugging Face mirror.

Multi-channel continuous background noise recordings from the 3rd CHiME Speech Separation and Recognition Challenge (CHiME-3), released under CC BY 4.0 so that they can be reused freely, including for commercial purposes.

Dataset scope: background noise recordings only. This mirror does not include the CHiME-3 speech utterances, transcriptions, annotations, or baseline software. The complete CHiME-3 distribution is catalogued by the Linguistic Data Consortium as LDC2017S24 and is subject to different licensing conditions.

102 WAV files: 17 recording sessions, each captured on all six microphone channels.

Environment	Description	Files
`BUS`	public bus	24
`CAF`	café	24
`PED`	outdoor pedestrian area	24
`STR`	street junction / traffic	30

Audio format: uncompressed WAV, 16-bit PCM, 16 kHz
Microphone configuration: six-channel tablet-mounted array; each channel (CH1–CH6) is a separate WAV file. There is no close-talking channel (CH0) in the background recordings.
Content: continuous environmental audio, used in CHiME-3 to construct simulated noisy-speech mixtures.

Directory layout

The original CHiME-3 layout is preserved so that existing data-preparation recipes work unchanged:

CHiME3/
├── README.md
├── checksums.md5
└── data/
    └── audio/
        └── 16kHz/
            └── backgrounds/
                ├── BGD_150203_010_CAF.CH1.wav
                ├── BGD_150203_010_CAF.CH2.wav
                └── ...

Filenames encode the recording session, the environment and the microphone channel. CHiME3/checksums.md5 is the file inventory; verify it from inside CHiME3/ with md5sum -c checksums.md5.

Usage

Download the whole tree:

hf download chimechallenge/chime3-background-noise --repo-type dataset --local-dir .

Or fetch a single environment:

hf download chimechallenge/chime3-background-noise --repo-type dataset \
  --include "CHiME3/data/audio/16kHz/backgrounds/*_CAF.*" --local-dir .

Licence

Creative Commons Attribution 4.0 International (CC BY 4.0), https://creativecommons.org/licenses/by/4.0/. You may copy, redistribute, modify and use these recordings for any purpose, including commercial use, provided appropriate attribution is given and modifications are indicated.

This licence covers the background-noise recordings only. Other components of the CHiME-3 corpus, including the LDC-distributed speech materials, may be licensed differently.

How to cite

Please cite both the CHiME-3 paper and the Zenodo record.

@inproceedings{barker2015third,
  author       = {Barker, Jon and Marxer, Ricard and Vincent, Emmanuel and Watanabe, Shinji},
  title        = {The third '{CHiME}' Speech Separation and Recognition Challenge: Dataset, task and baselines},
  booktitle    = {IEEE Automatic Speech Recognition and Understanding Workshop (ASRU)},
  pages        = {504--511},
  year         = {2015},
  organization = {IEEE}
}

@dataset{barker2026chime3backgrounds,
  author    = {Barker, Jon and Marxer, Ricard and Vincent, Emmanuel and Watanabe, Shinji},
  title     = {{CHiME-3} Multichannel Background Noise Recordings},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23571515},
  url       = {https://doi.org/10.5281/zenodo.23571515}
}