Speech Recognition Enhancement Via Robust CHMM Speech Background Discrimination

1 Abstract This paper describes a methodology for building robust Hidden Markov Model (HMM) based speech recognition system. The ultimate goal is to build a reliable large vocabulary isolated words speech recogniser. The model, that we are dealing with, is of continuous HMM type (CHMM). The topology selected is the left-right one as it is quite successful in speech recognition due to its consistency with the natural way of articulating the spoken words. One important task here is to efficiently extract the spoken words from their background using 3 states CHMM and process them in isolation by another 9 states models. This is considered as a perceptual way of extracting the signal. This technique is substantially increasing the performance of the system and improving the incorporation of states’ duration.