Orthogonalized Distinctive Phonetic Feature Extraction for Noise-Robust Automatic Speech Recognition