A General Algorithm for Training Hidden Markov Models with Multiple Observations
Xinmin Wang · Xiaogan Xueyuan xuebao · 2002
Hidden Markov models (HMMs) are statistical models capable of learning.They have been used successfully in many applications,especially for speech recognition.The classical BaumWelch algorithm assumes that all observations are independent of each other,but this is not the real case.This paper proposes a theoretical justification of the multiple observations HMMs training algorithm that does not impose the observation independence assumption and shows that the traditional BaumWelch algorithm is only a special case of observation independence assumption.