Language recognition by means of ergodic hidden Markov models

Abstract

A model, based on ergodic HMM's, is proposed for language recognition. This model, which can be trained without user intervention, in addition to modelling the sounds present in a specific language, also attempts to capture the typical combinations of sounds specific to that language. It is also s h o w n how this model can be extended to include a wider context than that offered b y a first order H M M w i t h o u t incurring the excessive computational burden of higher order Markov models.