作者: Wu Chou
DOI: 10.1121/1.426679
关键词: Artificial intelligence 、 Natural language processing 、 Phrase 、 Set (abstract data type) 、 Utterance 、 Key (cryptography) 、 State (computer science) 、 Sentence 、 Sequence 、 Computer science
摘要: A key-phrase detection and verification method that can be advantageously used to realize understanding of flexible ( i.e., unconstrained) speech. "multiple pass" procedure is applied to a spoken utterance comprising sequence words a "sentence"). First, plurality key-phrases are detected (i.e., recognized) based on a set phrase sub-grammars which may, for example, specific the state the dialogue. These then verified by assigning confidence measures thereto and comparing these threshold, resulting in set verified key-phrase candidates. Next, candidates connected into sentence hypotheses upon predetermined (e.g., task-specific) semantic information. And, finally, one or more sentence hypotheses are produce hypothesis and, from that, resultant understanding utterance.