A Parallel HMM Approach to Speech Recognition