In this paper, we describe a new word spotting algorithm that uses forward decoding. This algorithm is based on the Ergodic HMM. It basically performs comparison for the score of the last state of any word HMM. Therefore, it is expected to obtain a higher performance than continuous dynamic programming. Results of experiments showed a 36 % word accuracy rate which is superior to that by Viterbi decoding.
In the future, we plan to make a comparison between this new spotting algorithm and past continuous dynamic programming. In addition, we will try to make a new continuous speech recognition algorithm that combines Viterbi decoding and forward decoding.