Processed Silent EMG-features

This repository provides the hand-crafted feature representations used in our paper, "Can LLMs Understand Unvoiced Speech? Exploring EMG-to-Text Conversion with LLMs", ACL, 2025. These features are derived from the silent-EMG dataset originally introduced in "Digital Voicing of Silent Speech" (EMNLP 2020, arXiv:2010.02960). The raw dataset can be accessed via Zenodo. For more details on feature extraction and experimental setup, please refer to our paper.