Datasets:

mjinsky
/

refined_meta_dataset

id string	metadata:transcript list	metadata:vad list
V00_S0038_I00000125_P0059A	[ { "words": [ { "word": "Nice.", "start": 24.355000000000018, "end": 24.63499999999999, "score": 0.514 } ], "start": 24.355000000000018, "end": 24.63499999999999, "transcript": "Nice." }, { "words": [ { "word": "Yeah.", "st...	[ { "start": 24.20999999999998, "end": 24.75 }, { "start": 39.02600000000001, "end": 39.853999999999985 }, { "start": 49.874000000000024, "end": 50.35000000000002 }, { "start": 54.129999999999995, "end": 54.605999999999995 }, { "start": 61.394000000000005, "end"...
V00_S0038_I00000125_P0060	[ { "words": [ { "word": "That's", "start": 0.9950000000000045, "end": 1.1750000000000114, "score": 0.7 }, { "word": "interesting.", "start": 1.1949999999999932, "end": 1.5749999999999886, "score": 0.833 } ], "start": ...	[ { "start": 1.1379999999999768, "end": 1.7420000000000186 }, { "start": 1.906000000000006, "end": 2.8940000000000055 }, { "start": 3.410000000000025, "end": 9.677999999999997 }, { "start": 9.906000000000006, "end": 10.382000000000005 }, { "start": 11.08999999999997...
V00_S0038_I00000126_P0059A	[ { "words": [ { "word": "Okay,", "start": 0.19000000000005457, "end": 0.4900000000000091, "score": 0.298 }, { "word": "okay.", "start": 1.11099999999999, "end": 1.4510000000000218, "score": 0.715 } ], "start": 0.19000...	[ { "start": 0.14599999999995816, "end": 0.6219999999999573 }, { "start": 1.0739999999999554, "end": 1.5499999999999545 }, { "start": 8.018000000000029, "end": 12.173999999999978 }, { "start": 13.426000000000045, "end": 18.605999999999995 }, { "start": 18.7699999999...
V00_S0038_I00000126_P0060	[ { "words": [ { "word": "OK.", "start": 0.006999999999948159, "end": 0.18700000000001182, "score": 0.744 } ], "start": 0.006999999999948159, "end": 0.18700000000001182, "transcript": "OK." }, { "words": [ { "word": "Do", "s...	[ { "start": 0, "end": 0.2699999999999818 }, { "start": 1.1380000000000337, "end": 3.6620000000000346 }, { "start": 12.018000000000029, "end": 12.42999999999995 }, { "start": 12.690000000000055, "end": 13.101999999999975 }, { "start": 18.51400000000001, "end": 1...
V00_S0038_I00000129_P0059A	[ { "words": [ { "word": "yeah.", "start": 53.317999999999984, "end": 53.4380000000001, "score": 0.188 } ], "start": 53.317999999999984, "end": 53.4380000000001, "transcript": "yeah." }, { "words": [ { "word": "Hmm.", "start...	[ { "start": 53.61799999999994, "end": 53.99800000000005 }, { "start": 136.04999999999995, "end": 136.3979999999999 }, { "start": 204.81799999999998, "end": 205.5820000000001 } ]
V00_S0038_I00000129_P0060	[ { "words": [ { "word": "Bye.", "start": 3.3450000000000273, "end": 3.505000000000109, "score": 0.503 } ], "start": 3.3450000000000273, "end": 3.505000000000109, "transcript": "Bye." }, { "words": [ { "word": "Okay.", "star...	[ { "start": 50.96199999999999, "end": 54.44599999999991 }, { "start": 59.922000000000025, "end": 60.26999999999998 }, { "start": 92.04999999999995, "end": 92.68599999999992 }, { "start": 93.106, "end": 93.67800000000011 }, { "start": 106.25800000000004, "end": ...
V00_S0038_I00000131_P0059A	[{"words":[{"word":"I","start":8.001999999999953,"end":8.10200000000009,"score":0.742},{"word":"bet.(...TRUNCATED)	[{"start":7.938000000000102,"end":8.57400000000007},{"start":33.153999999999996,"end":33.59799999999(...TRUNCATED)
V00_S0038_I00000131_P0060	[{"words":[{"word":"I'm","start":0.0,"end":0.08200000000010732,"score":0.748},{"word":"the","start":(...TRUNCATED)	[{"start":0.0,"end":1.3740000000000236},{"start":1.7940000000000964,"end":4.574000000000069},{"start(...TRUNCATED)
V00_S0038_I00000132_P0059A	[{"words":[{"word":"Okay.","start":0.0,"end":0.1310000000000855,"score":0.623}],"start":0.0,"end":0.(...TRUNCATED)	[{"start":0.0,"end":0.39799999999991087},{"start":1.2980000000000018,"end":1.8699999999998909},{"sta(...TRUNCATED)
V00_S0038_I00000132_P0060	[{"words":[{"word":"Yeah.","start":4.15300000000002,"end":4.393000000000029,"score":0.464}],"start":(...TRUNCATED)	[{"start":4.114000000000033,"end":4.461999999999989},{"start":5.490000000000009,"end":6.701999999999(...TRUNCATED)

End of preview.

refined_meta_dataset

A processed subset derived from Meta's Seamless Interaction dataset.

This release keeps the original multimodal media and adds per-sample SMPL-X motion estimates produced by our cleaning/preprocessing pipeline (based on GVHMR), plus 50fps resampled motion, HumanML3D-style joint features, and voice-conversion tokens.

Provenance & changes

Source dataset: facebook/seamless-interaction
Source splits used here:
- improvised/train/{0000,0001,0002,0003,0004,0005,0006,0007,0008,0009,0010}
- improvised/dev/{0000,0001}
- improvised/test/{0000}
What we changed / added:
- Ran a cleaning/preprocessing pipeline on selected samples
- Added *_smplx.npz (SMPL-X parameters, original fps)
- Added *_50fps_smplx.npz (SMPL-X resampled to 50 fps)
- Added *_50fps_joints_xyz.npy and HumanML3D-style new_joints/ / new_joints_vecs/
- Added voice_conversion_tokens.pt for most samples
- Kept accompanying .mp4, .wav, and .json metadata for each sample
This is not an official Meta release.

Structure

improvised/{train,dev,test}/{shard_id}/{sample_id}/
  - {sample_id}.mp4
  - {sample_id}.wav
  - {sample_id}.json
  - {sample_id}_smplx.npz
  - {sample_id}_50fps_smplx.npz
  - {sample_id}_50fps_joints_xyz.npy
  - new_joints/{sample_id}_50fps_joints_xyz.npy
  - new_joints_vecs/{sample_id}_50fps_joints_xyz.npy
  - voice_conversion_tokens.pt   # present for most samples

Train, dev, and test shards are stored under separate directories and should not be mixed.

License

Because this dataset adapts Seamless Interaction material, it is released under the same license:

CC-BY-NC 4.0

You must:

give appropriate credit to Seamless Interaction / Meta;
indicate that changes were made;
not use the material for commercial purposes without permission.

Please also review the original dataset card and usage policy: https://huggingface.co/datasets/facebook/seamless-interaction

Citation

If you use this dataset, please cite Seamless Interaction (and this derivative if relevant):

@inproceedings{seamless2024,
  title     = {Seamless Interaction},
  author    = {{Meta FAIR}},
  year      = {2024},
  url       = {https://huggingface.co/datasets/facebook/seamless-interaction}
}

Note: replace/update the citation with the official paper citation from the Seamless Interaction README when available.

Downloads last month: 14

Total file size:

835 GB