The paper proposes a new speech recognition system based on CSS (client and server system) architecture and describes a telephone application which combines an existing telephone PBX system and an OA (office automation) system consisting of personal computers and fax machines, The server, which is separated from application software, is mainly for speech recognition, and the client is an application-driven processor which includes a speech input part and application software. The CSS-based speech recognition system makes it much easier to use the installed speech recognition server to various applications without any big changes of the processing algorithm and architecture. The authors also propose a new method for making acoustic models which cover speakers' speech feature variety. This method is based on an automatic generation of speech recognition units from a large speech database.
展开▼
