A HMM-based approach for segmenting continuous speech

Several algorithms used for automatically segmenting an input speech signal are reviewed. It is shown that they either incorporate noise as a part of the word to be enrolled or falsely classify a portion of a word as noise. As a result, recognition performance suffers. Another approach to automatically segmentating continuous speech is presented. To verify this approach, experimental results from a database of 30 speakers whose speech has been recorded over the public switched telephone network are presented. The results benchmark the algorithm against a state-of-the-art approach and show a 4× reduction in the error rate of the recognition system