A fast decoding algorithm based on sequential detection of the changes in distribution

A fast algorithm for left-to-right HMM decoding is proposed in this paper. The algorithm is developed based on a sequential detection scheme which is asymptotically optimal in the sense of detecting a possible change in distribution as reliably and quickly as possible. The scheme is extended to HMM decoding in determining the state segmentations for likelihood or other score computations. As a sequential scheme, it can determine a state boundary in a few time steps after it occurs. The examples in this paper show that the proposed algorithm is 5 to 9 times faster than the Viterbi algorithm while it still can provide the same or similar decoding results. The proposed algorithm can be applied to speaker recognition, audio segmentation, voice/silence detection, and many other applications, where an assumption of the algorithm is usually satisfied.