Feature-Dependent Allophone Clustering

摘要: We propose a novel method for clustering allophones called Feature-Dependent Allophone Clustering (FD-AC) that determines feature-dependent HMM topology automatically. Existing methods allophone are based on parameter sharing between the models resemble each other in behaviors of feature vector sequences. However, all features sequences may not necessarily have common structures. It is considered can be better modeled by allocating optimal structure to feature. In this paper, we Successive State Splitting (FD-SSS) as an implementation FD-AC. speaker-dependent continuous phoneme recognition experiments, HMMs created FD-SSS reduced error rates about 10% compared with conventional features.