A new method for fuzzy hidden Markov models in speech recognition

Abstract

Thispaperproposes a Ftuzzi approach tothe Hidden Markov Model(HMM). This method called the fizzy HMMfbrspeech andspeaker recognition asan application offizzy expectation maximizing algorithm inHMM.Thefitzzy HMM algorithm isregar-ded asan application of'the,fizzy expectation-maximization (EM)algorithm totheBatum-Welch algorithm inthe HMM. TheTexas Instruments p4uised speech and speaker recognition experiments andshowbetter results fbr.fiuzzv HMMscompared withconventional HMMs.Equation andhowestimation ofdiscrete and continuious HMM parameters on basedthistwo algorithm isexplained andperfbrmance oftwospeech recognition methodtbronehundred issurveved. This papershowbetter results forthefiuzzy HMVM, compared with theconventional HMM.