We introduce a large-scale multilingual lip-reading dataset: MultiVSR. The dataset comprises a total of 12,000 hours of video footage, covering English + 12 non-English languages. MultiVSR is a massive dataset with a huge diversity in terms of the speakers as well as languages, with approximately 1.6M video clips across 123K YouTube videos. Please check the website for samples.
Please check the GitHub repo to download… See the full description on the dataset page:
https://huggingface.co/datasets/sindhuhegde/multivsr.
