AVE Speech: A Comprehensive Multi-Modal Dataset for Speech Recognition Integrating Audio, Visual, and Electromyographic Signals
Abstract
AVE Speech is a large-scale Mandarin speech corpus that pairs synchronized audio, lip video and surface electromyography (EMG) recordings. The dataset contains 100 sentences read by 100 native speakers. Each participant repeated the full corpus ten times, yielding over 55 hours of data per modality. These complementary signals enable… See the full description on the dataset page: https://huggingface.co/datasets/MML-Group/AVE-Speech.
