グラフを利用して誤りを含む記号列を分類する方法とその音声認識への応用

概要

論文の詳細を見る
Classification of symbol strings is one of the important problems in pattern recognition. For this Problem, usually the input string is compared with some reference strings and identified with the reference string having the highest similarity. But when the input string contains some errors or some extra symbols, its classification becomes difficult. We propose a method which overcomes this difficulty by using graph theory, and explain the algorithm for its performance and verify the completeness of the algorithm. Next, we have applied this method to speech recognition and obtained good results.By comparing with another method using dynamic programming (so-called DP method), it has become clear that in our method it takes a little more time to classify input samples than the DP method.but when input samples contain some errors or extra symbols, the classification score of our method is much better than the DP method.
一般社団法人情報処理学会の論文
1978-09-15