This paper describes the connected words recognition experiment using Continuous Dynamic Programming (CDP) and an analysis of the experimental result. Continuous DP is a word-spotting method which uses the similarity of each word frame wisely and the similarity (CDP value) is the time warping distance normalized by the length of the warping path. A procedure is proposed to determine words using the word lattice from CDP values and recognition experiments are performed. The relation between the vocabulary size and the recognition rate, the separability of word lattice, comparison with the isolated word recognition, and its error tendency were checked experimentally. The following recognition rate for speaker-dependent recognition of connected words was obtained: 93. 8%, 92. 2%, 90. 0%, and 88. 8% for the vocabulary size of 100, 200, 300, and 400.
|ジャーナル||Systems, computers, controls|
|出版ステータス||Published - 1984 7月|
ASJC Scopus subject areas