An efficient Korean speech recognition for name and digit on a communication network

Abstract

The necessity of service using speech recognition goes on increasing day by day, but for the difficulty of speech recognition, the efficient application service is not activated yet. For the security of users and the convenient use of a disabled person, a desire for using speech about the user’s name, ID number, or an account which is mainly used in service is increasing accordingly as the information and banking service on communication network increases. This paper presents the efficient Korean speech recognition method for name and digit string on a communication network. Because most of the traditional methods recognize the full name at once, there are some problems, in that they must have all patterns of the name or require very accurate segmentation of the phoneme, etc. So the unit of Korean spelling pronunciation that is contained in the name is recognized, and then a connection is made to the full name recognition. In the recognition of the digit, semisyllable and the combination of a connected semisyllable are used. That is, 20 semisyllables from 0 to 9 are used as the pattern of the start semisyllable or the end semisyllable of the digit string in recognition, and 100 connected semisyllables from 00 to 99 are used as the pattern of the connection part of each digit string. In comparison with the existing method, all names and digits, regardless of their kind and figure can be recognized.