A hybrid syllable recognition system based on vowel spotting

摘要: In this paper we present a hybrid ANN/HMM syllable recognition system based on vowel spotting. Using an advanced multilevel vowel-spotting module track all phonemes in speech signals from where model the segments located between two successive vowels which are defined as syllables. order to achieve minimum losses and accurate detection, focus taking special care of spotter is three different techniques: discrete hidden Markov models (DHMMs), multilayer perceptrons heuristic rules.To set up segments, DHMMs with multiple codebooks used. The usual DHMM probability parameters replaced by combined neural network outputs. For purpose, use both context dependent independent networks.The was tested TIMIT NTIMIT databases results obtained showed 75.09% 59.30% average accuracy, respectively. It has be noted that above no grammars or syllable-based lexicons were