Speech recognition method based on weighed autoregressive HMM

For non-independent speech recognition, in order to solve the problem of the assumption that the observation vectors are independent and the amount of data is small in Hidden Markov Model, a weighted autoregressive Hidden Markov Model was presented based on the Continuous Hidden Markov Model in this paper. The weighted autoregressive process was exploited to extract the observation vector, which is more suitable for recognition of the actual voice signals with strong random characteristic.