CHiME-3 Multichannel Background Noise Recordings

CHiME-3 Multichannel Background Noise Recordings This dataset contains the multichannel background-noise recordings collected for the 3rd CHiME Speech Separation and Recognition Challenge (CHiME-3). These recordings were made in four real-world acoustic environments: BUS: Public transport (bus) CAF: Café PED: Pedestrian areas STR: Streets with traffic noise The recordings were captured using a six-channel microphone array mounted on a tablet device. Audio is provided as uncompressed WAV files, sampled at 16 kHz with 16-bit resolution. Dataset Scope This deposit contains only the continuous background-noise recordings from CHiME-3. It does not include the speech utterances, annotations, or baseline software. The recordings retain the directory structure and filenames used in the original CHiME-3 distribution, allowing them to be used with existing CHiME-3 data preparation and evaluation pipelines. The complete CHiME-3 corpus, including speech recordings and associated resources, is available through the Linguistic Data Consortium (LDC2017S24). Downloading and Reconstructing the Archive To facilitate reliable uploading and downloading of this large dataset, the original tar archive has been divided into smaller files, each approximately 1 GB. To reconstruct the archive: Download all files named CHiME3-backgrounds.tar.part-* into the same directory. Open a terminal in that directory. Reassemble the archive using: cat CHiME3-backgrounds.tar.part-* > CHiME3-backgrounds.tar Extract the reconstructed archive: tar -xf CHiME3-backgrounds.tar The reconstruction is lossless: concatenating the parts in filename order produces the original tar archive exactly. The extracted files can then be used directly with CHiME-3-compatible tools and scripts. After reconstruction, the md5sum checksum should be as follows: ❯ md5sum CHiME3-backgrounds.tar37a562aa8ce9e97413e55a9037f0c51f CHiME3-backgrounds.tar Note on Archive Extraction Warnings When extracting this archive on Linux using standard GNU tar, you may see warnings such as: tar: Ignoring unknown extended header keyword 'LIBARCHIVE.xattr...' These are harmless macOS filesystem metadata tags (com.apple.provenance, com.apple.macl) embedded during archive creation. They can be safely ignored and do not affect the extracted files or data integrity. To extract without showing these warnings on Linux: tar --warning=no-unknown-keyword -xvf archive.tar.gz (Note: macOS users running default tar -xvf will not encounter these warnings.) Licence These background-noise recordings are released under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence, permitting reuse and redistribution, including for commercial purposes, with appropriate attribution. Citation Users of these recordings are requested to cite both the original CHiME-3 publication and this Zenodo dataset record. Barker, J., Marxer, R., Vincent, E., & Watanabe, S. (2015). The third 'CHiME' Speech Separation and Recognition Challenge: Dataset, task and baselines. Proceedings of IEEE ASRU 2015, pp. 504–511. Please also cite this Zenodo deposit using the DOI provided in the record's citation information.