chime6 / README.md
jonbarker68's picture
Create README.md
efe5ac2 verified
|
Raw History Blame Contribute Delete
3.85 kB
metadata
license: cc-by-4.0
viewer: false
language:
  - en
tags:
  - audio
  - speech-recognition
  - speech-separation
  - chime
size_categories:
  - n<1K

CHiME-6 Dataset

CHiME-6 is a dataset for speech separation and recognition in everyday home environments. It contains recordings from 20 dinner parties, each held in a different home. For details on the recording setup, data formats, and file naming, see the official CHiME-5 data documentation.

The dataset was originally used in the CHiME-6 Challenge in 2020 and in the CHiME-7 DASR task in 2023. It is derived from CHiME-5 data by fixing clock-drift synchronization issues across the recording devices (six 4-channel Microsoft Kinect arrays and four binaural microphones worn by the participants) and providing updated force-aligned transcriptions and official train/dev/eval splits for distant multi-microphone conversational speech recognition.

Dataset Details

  • Audio: ~50 hours of multi-channel conversational speech (16 kHz, 16-bit WAV) recorded across 20 sessions (16 train, 2 dev, 2 eval) in real domestic kitchens, dining rooms, and living rooms.

  • Microphones: 6 Kinect microphone arrays (24 distant channels total) plus 4 binaural close-talking microphone pairs worn by the four participants in each session.

  • Annotations: Time-aligned orthographic transcriptions with speaker IDs and session metadata in JSON format, including official utterance segmentations for both Track 1 (ASR) and Track 2 (Diarization + ASR).

Directory layout

The Hugging Face repository contains five downloadable archives under data/. Each archive extracts into a top-level CHiME6/ directory. Extract the archives you need into the same parent directory to combine their contents under CHiME5/.

chime6/
├── README.md
├── LICENSE.txt
└── data/
    ├── CHiME6_train.tar.gz
    ├── CHiME6_dev.tar.gz
    ├── CHiME6_eval.tar.gz
    ├── CHiME6_transcriptions.tar.gz
    └── CHiME6_floorplans.tar.gz

Usage

Download all five archives:

hf download chimechallenge/chime6 --repo-type dataset \
  --include "data/*.tar.gz" --local-dir .

Or download only the development archive:

hf download chimechallenge/chime6 data/CHiME6_dev.tar.gz \
  --repo-type dataset --local-dir .

Extract the development archive to create ./CHiME6/:

tar -xzf data/CHiME6_dev.tar.gz

Run the same extraction command for each additional archive from this directory.

Licence

This release is licensed under the Creative Commons Attribution 4.0 International licence. See LICENSE.txt for the licence text and attribution information.

Citation

If you use CHiME-6, please cite the following papers:

@inproceedings{barker18_interspeech,
  author    = {Jon Barker and Shinji Watanabe and Emmanuel Vincent and Jan Trmal},
  title     = {The Fifth {CHiME} Speech Separation and Recognition Challenge: Dataset, Task and Baselines},
  booktitle = {Proceedings of Interspeech 2018},
  year      = {2018},
  pages     = {1561--1565},
  doi       = {10.21437/Interspeech.2018-1768}
}

@inproceedings{watanabe2020chime,
  title={CHiME-6 Challenge: Tackling multispeaker speech recognition for unsegmented recordings},
  author={Watanabe, Shinji and Mandel, Michael and Barker, Jon and Vincent, Emmanuel and Arora, Ashish and Chang, Xuankai and Khudanpur,
 Sanjeev and Manohar, Vimal and Povey, Daniel and Raj, Desh and others},
  booktitle={CHiME 2020-6th International Workshop on Speech Processing in Everyday Environments},
  year={2020}
}

Links