File size: 2,792 Bytes
03c6686
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
595f6ba
03c6686
 
 
 
595f6ba
 
03c6686
 
 
 
 
 
 
 
 
 
 
595f6ba
03c6686
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: other
library_name: pytorch
tags:
  - co-speech-gesture-generation
  - co-speech-motion-generation
  - speech-driven-motion
  - human-motion-generation
  - masked-modeling
  - beat2
  - smpl-x
  - echomask
datasets:
  - H-Liu1997/BEAT2
---

# EchoMask official checkpoints

Official checkpoints for **EchoMask: Speech-Queried Attention-based Mask Modeling for Holistic Co-Speech Motion Generation** (ACM Multimedia 2025), a speech-conditioned approach to holistic **co-speech gesture generation**.

- [Paper](https://arxiv.org/abs/2504.09209) · [Hugging Face Paper page](https://huggingface.co/papers/2504.09209)
- [Project page](https://xiangyuezhang.com/EchoMask/) · [Code](https://github.com/Xiangyue-Zhang/EchoMask)
- [Published version](https://doi.org/10.1145/3746027.3754847)
- [Generated inference data](https://huggingface.co/datasets/X-Zhang/EchoMask-Inference-Data)

## Files

| Archive | Protocol | Contents |
| --- | --- | --- |
| `weights_echomask_no_smplx.zip` | Speaker 2, paper protocol | Original pretrained representation models and EchoMask checkpoint; SMPL-X files excluded |
| `EchoMask_all_speakers_weights_25spk.zip` | 25 English BEAT2 speakers | Reusable representation models and selected EchoMask checkpoint |

The two archives implement different training protocols and should not be treated as a controlled single-speaker versus multi-speaker ablation. See the code repository for installation paths and configuration details.

Verify both downloads with `sha256sum -c SHA256SUMS`.

## Reported results

| Protocol | Speakers | FGD ↓ | BC ↑ | DIV ↑ | MSE ↓ | LVD ↓ |
| --- | ---: | ---: | ---: | ---: | ---: | ---: |
| Speaker 2 (paper) | 1 | 0.4623 | 0.7738 | 13.370 | 6.761e-8 | 7.290e-5 |
| Released all-speaker checkpoint | 25 | 0.5656 | 0.4951 | 9.299 | 4.700e-8 | 6.090e-5 |

The first row is from the paper. The second row is from the released all-speaker checkpoint. Their protocols differ.

## Terms

This repository mirrors research artifacts released by the authors. No new license is granted by this model card. The Hugging Face archives intentionally exclude `SMPLX_NEUTRAL_2020.npz`; obtain SMPL-X directly from the [official source](https://smpl-x.is.tue.mpg.de/) under its terms. Code, BEAT2, pretrained encoders, and other third-party assets remain subject to their respective terms.

## Citation

```bibtex
@inproceedings{zhang2025echomask,
  title={EchoMask: Speech-Queried Attention-based Mask Modeling for Holistic Co-Speech Motion Generation},
  author={Zhang, Xiangyue and Li, Jianfang and Zhang, Jiaxu and Ren, Jianqiang and Bo, Liefeng and Tu, Zhigang},
  booktitle={Proceedings of the 33rd ACM International Conference on Multimedia},
  pages={10827--10836},
  year={2025},
  doi={10.1145/3746027.3754847}
}
```