High-Accuracy Large-Vocabulary Speech Recognition Using Mixture Tying and Consistency Modeling

13166

Knowledge Triples