File size: 4,251 Bytes
446eb7c
 
71c0101
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
446eb7c
71c0101
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d81f0cd
71c0101
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
---
license: apache-2.0
language:
  - en
tags:
  - co-speech-gesture
  - smplx
  - smplx-flame
  - beat2
  - beatx
  - miburi
  - rag-gesture
  - discourse-relations
  - prominence
  - whisper-transcription
size_categories:
  - 1K<n<10K
pretty_name: "BEAT2-English + additional annotations (RAG-Gesture / MIBURI)"
---

# BEAT2 Official Release + Additional Annotations

This is a fork of [`H-Liu1997/BEAT2`](https://huggingface.co/datasets/H-Liu1997/BEAT2)
that adds annotations contributed by the
[**RAG-Gesture**](https://vcai.mpi-inf.mpg.de/projects/RAG-Gesture/) (CVPR 2025)
and [**MIBURI**](https://vcai.mpi-inf.mpg.de/projects/MIBURI/) (CVPR 2026) projects.
The base BEAT2-English data (motion, audio, TextGrids, semantic labels,
pretrained motion-autoencoder weights) is inherited verbatim from upstream;
the additional annotations from RAG-Gesture and MIBURI are pushed on top.

## Citations

If you use **only** the original BEAT2 dataset, please cite the original BEAT/EMAGE paper:

```bibtex
@inproceedings{liu2024emage,
  title={Emage: Towards unified holistic co-speech gesture generation via expressive masked audio gesture modeling},
  author={Liu, Haiyang and Zhu, Zihao and Becherini, Giorgio and Peng, Yichen and Su, Mingyang and Zhou, You and Zhe, Xuefei and Iwamoto, Naoya and Zheng, Bo and Black, Michael J},
  booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
  pages={1144--1154},
  year={2024}
}
```

If you use the **additional annotations of discourse relations** (`discourse_rels/`)
or **prominence** (`prom/`), please also cite RAG-Gesture:

```bibtex
@inproceedings{mughal2025retrieving,
  title={Retrieving semantics from the deep: an rag solution for gesture synthesis},
  author={Mughal, M Hamza and Dabral, Rishabh and Scholman, Merel CJ and Demberg, Vera and Theobalt, Christian},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={16578--16588},
  year={2025}
}
```

If you use the **25 fps SMPL-X + FLAME** (`smplxflame_25/`), please consider citing MIBURI:

```bibtex
@InProceedings{mughal2026miburi,
    title={MIBURI: Towards Expressive Interactive Gesture Synthesis},
    author={M. Hamza Mughal and Rishabh Dabral and Vera Demberg and Christian Theobalt},
    booktitle={Computer Vision and Pattern Recognition (CVPR)},
    year={2026}
}
```

## Contents

Original BEAT2-English dataset (inherited from upstream `H-Liu1997/BEAT2`):

- `smplxflame_30/` — SMPL-X body + FLAME expressions at 30 fps
- `textgrid/` — phoneme + word TextGrid alignments
- `wave16k/` — 16 kHz mono audio
- emotion label
- `sem/` — semantic annotation
- `weights/` — pretrained CNN-based motion autoencoders for BEAT (used by the FGD metric)

New data annotations by the RAG-Gesture & MIBURI projects (pushed in this fork):

- `discourse_rels/` — discourse relations (RAG-Gesture)
- `prom/` — prominence values (RAG-Gesture)
- `smplxflame_25/` — 25 fps version of the 30 fps SMPL-X + FLAME body (MIBURI)
- `whisper_transcription/` — Whisper transcripts that recover casing and
  punctuation that the upstream TextGrid normalization removed (MIBURI)

## License

This repository carries the upstream **Apache 2.0** license from
[`H-Liu1997/BEAT2`](https://huggingface.co/datasets/H-Liu1997/BEAT2). The
RAG-Gesture and MIBURI annotations are released under the same Apache 2.0
license for ease of downstream reuse.

## Missing Data

These takes are missing from the original BEAT2-English release; the fork
inherits that gap:

```
9_miranda_0_1_8
15_carlos_0_6_6, 15_carlos_0_12_12
21_ayana_0_1_8
6_carla_0_1_48
23_hailing_0_73_74
25_goto_0_1_1, 25_goto_0_5_5
```

## Speaker Name

```python
lut = {
    "wayne": 1,
    "scott": 2,
    "kieks": 10,
    "nidal": 11,
    "lu": 13,
    "zhao": 12,
    # "zhang": 14,
    "carlos": 15,
    "jorge": 16,
    "itoi": 17,
    "daiki": 18,
    # "jaime": 19,
    "li": 20,
    "ayana": 21,
    "luqi": 22,
    "hailing": 23,
    "kexin": 24,
    "goto": 25,
    # "reamey": 26,
    "yingqing": 27,
    "tiffnay": 28,
    # "hanieh": 29,
    "katya": 30,
    "solomon": 3,
    "lawrence": 4,
    "stewart": 5,
    "carla": 6,
    "sophie": 7,
    # "catherine": 8,
    "miranda": 9,
}
```