Dataset · academic5000
Chinese announcer speech video dataset for gesture generation
academic5000/CAS
Over 19,000 videos of Chinese announcers for training and evaluating portrait video generation models with gestures.
What's inside
The Chinese Announcer Speech (CAS) dataset was prepared by authors of a scientific article for the TCSVT 2026 journal. It includes over 19,000 Chinese-language video clips featuring more than 800 unique speakers and is designed to evaluate portrait video generation where gestures match speech.
What you can do with it
- Training and validation of talking avatar and speech-driven gesture generation models
- Researching the synchronization of hand and body movements with Chinese speech streams
- Languages
- Chinese
- Size
- 10K–100K rows
- Downloads this month
- 561
- Likes
- 4
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("academic5000/CAS")