Double Layer Architectures for Automatic Speech Recognition Using HMM

Overcoming HMM Time and Parameter Independence Assumptions for ASR

“…In an effort to model the temporal properties of the speech signal, class labels modeling (Stemmer et al, 2003) has been studied in a double layer speech recognition framework (Casar & Fonollosa, 2006a). The main idea was to deal with acoustic and temporal information in two different steps.…”