| license: other | |
| pretty_name: SemTalk Inference Data | |
| language: | |
| - en | |
| tags: | |
| - co-speech-gesture-generation | |
| - motion-generation | |
| - beat2 | |
| - smpl-x | |
| size_categories: | |
| - n<1K | |
| # SemTalk Inference Data | |
| Generated SMPL-X predictions from the official SemTalk checkpoints for comparison and visualization without rerunning the full generation pipeline. | |
| - [Paper](https://arxiv.org/abs/2412.16563) · [Hugging Face Paper page](https://huggingface.co/papers/2412.16563) | |
| - [Project page](https://xiangyuezhang.com/SemTalk/) · [Code](https://github.com/Xiangyue-Zhang/SemTalk) | |
| - [Model checkpoints](https://huggingface.co/X-Zhang/SemTalk) | |
| ## Files | |
| | Archive | Protocol | Description | | |
| | --- | --- | --- | | |
| | `SemTalk_testnpz.zip` | Speaker 2, paper protocol | Generated test `.npz` files | | |
| | `SemTalk_all_speakers_results_npz_25spk.zip` | 25 English BEAT2 speakers | Generated test `.npz` files | | |
| Verify both downloads with `sha256sum -c SHA256SUMS`. | |
| The two releases use different training protocols and should not be interpreted as a controlled single-speaker versus multi-speaker ablation. | |
| ## Provenance and terms | |
| These archives contain generated predictions, not BEAT2 ground-truth motion or audio. Users remain responsible for complying with the licenses and terms of [BEAT2](https://huggingface.co/datasets/H-Liu1997/BEAT2), [SMPL-X](https://smpl-x.is.tue.mpg.de/), and the SemTalk repository. This card grants no additional rights. | |