System for Temporal Registration of Quasi Phonemic Utterance Representations
Original Publication Date: 1980-Dec-01
Included in the Prior Art Database: 2005-Feb-14
A method is described for temporal registration of two utterances for purposes of testing for a match between an unknown utterance and a (set of) reference(s), based upon preliminary segmentation into states corresponding, in most cases, to phoneme classes (fricative elements, vowel classes, nasal classes, etc.). Classification at the phonetic state in speech sound is performed prior to time alignment. In the first stage, phonetic (sound) classes are assigned, and in the second stage the phonetic classes are used in dynamic time warping to perform temporal registration.