Researchers have introduced VSRo-200, a new large-scale dataset for visual speech recognition in Romanian. The dataset includes 200 hours of real-world podcast videos, with a portion annotated by humans and the rest by pseudo-labels from a fine-tuned Romanian ASR model. This resource aims to establish a benchmark for low-resource visual speech recognition and facilitate studies on supervision quality, domain generalization, and multimodal fusion. AI
IMPACT Provides a new resource for advancing research in low-resource visual speech recognition and multimodal fusion.
RANK_REASON The cluster describes a new academic dataset and benchmark for a specific research area (visual speech recognition).
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
