The Method of Words Phonetic Decoding Using Kullback- Leibler Information Discrimination for High-Speed Performance Systems of Automatic Speech Analysis and Recognition
Keywords:
Automatic Speech Recognition, Pattern Recognition, Supervised Learning, Minimum Information Discrimination CriterionAbstract
There has been presented a new kind of words phonetic decoding for a limited set of minimal speech units (separate phonemes) as an alternative to known speech recognition methods based on hidden Markov models of speech signals. It is based on an idea of multiple (by an order of magnitude or more) data compression due to mapping of words and phrases from a vocabulary to a sequence of phonetic codes. The achieved effect confirmed by results of experimental researches is an increase in computational speed of speech signals while preserving adequate speech recognition accuracy and reliability.