Speech understanding system using knowledge engineering techniques

This paper introduces the outline of our speech understanding system. The overall flow of control is based on the bottom-up processing. The simulation system is under construction and consists of the hierarchically connected acoustic and lexical analyzers. The word recognition based on the phoneme recognition is performed as the main process of the acoustic analyzer. A continuous DP matching technique is applied for the connected digit recognition, after the lexical analyzer appoints the section that may contain a sequence of digits. The lexical analyzer consists of a parser and a discourse manager. The parser scans the input words in a left-to-right fashion, getting the grammatical and semantic rules of each key word from the dictionary entry. Then the connectivity of these key words are checked using the extracted rules. A verification on the existence of a digit sequence or a functional word between two milestone words is done at this stage. The output of a parser are filtered out by the discourse manager to choose a possible sentence under each context. The current task domain is the reception of international telephone call.