Proposal of Ladder Continuous DP for Connected Pattern Spotting Applied to Sentence Spotting

摘要

Human speech includes quite a lot of sentences which are out of task domain. It is thought that spotting intent or sentences represents an effective approach for human-machine interface. In previous work, we developed connected word(sentence) recognition systems. This paper proposes the Ladder CDP(LCDP) method, a new connected pattern spotting algorithm that is very simple in structure. The new method is not just confined to speech, for it is also able to integrate gesture spotting for multi-modal interface. Results for sentence spotting experiments showed the present method was confirmed to be effective, compared with previous works.