ground-truth/multichannel-meetings-10h

GroundTruth Multi-Channel Meeting Audio Dataset (10h)

Summary

This dataset contains approximately 10 hours of co-located, multi-speaker meeting recordings, each captured simultaneously via a room (built-in) microphone and individual close-talk lapel microphones worn by each participant.

Each meeting includes:

• One full meeting recording (room microphone)
• Individual close-talk recordings for each participant (one file per speaker)
• Structured metadata describing speakers, environment, and recording setup

The dataset is designed to support development of robust multi-speaker audio systems under realistic acoustic conditions.

Key Features

• Dual-modality capture: room recording plus per-speaker close-talk tracks per session
• Complete speaker coverage: every participant has an isolated close-talk file
• 12 unique speakers reused across meetings with consistent speaker IDs
• Controlled acoustic environments: co-located participants in small rooms
• Rich metadata per session and per speaker

Dataset Characteristics

• Total duration: about 10 hours (36,289 seconds)
• Total meetings: 40
• Unique speakers: 12
• Speakers per meeting: 3 (fixed)
• Audio format: .m4a
• Language: English (en-IN, Indian English)
• Recording date: March 2026

Loading the Dataset

from datasets import load_dataset

ds = load_dataset("ground-truth/multichannel-meetings-10h", split="train")
print(ds[0].keys())

Row-level fields in the published dataset (dataset_index.json):

• meeting_id
• domain
• full_audio_path (path to room microphone audio file)
• speaker_audio_paths (list of close-talk speaker audio paths)
• speaker_ids
• num_speakers
• duration_sec
• metadata_path (path to JSON file containing meeting_metadata and speakers_metadata)

If you want decoded Audio columns (room_audio, speaker_audios) plus expanded metadata fields, use scripts/build_dataset.py.

Use Cases

• Speaker diarization
• Multi-speaker ASR
• Speech source separation

License

This dataset is released under CC BY-NC 4.0. It may be used for non-commercial research and evaluation purposes with attribution.

Citation

@dataset{groundtruth_multichannel_meetings_10h,
  title   = {GroundTruth Multi-Channel Meeting Audio Dataset (10h)},
  author  = {GroundTruth},
  year    = {2026},
  license = {CC-BY-NC-4.0},
  url     = {https://huggingface.co/datasets/ground-truth/multichannel-meetings-10h}
}

Contact

• Hugging Face org: ground-truth
• Website: ground-truth.in
• Email: hello@ground-truth.in