Advances in Missing Feature Techniques for Robust Large-Vocabulary Continuous Speech Recognition