Designing Practical Models for Isolated Word Visual Speech Recognition