| Summary: | 碩士 === 國立清華大學 === 資訊工程學系 === 90 === Speech dialogue is the most intuitive presentation of computer system user interface. We propose a dialogue interface agent for the user interface agent of service agent systems. It separates the interface from service systems. The dialogue interface agent guides users to use service systems. In this thesis, the dialogue interface agent interacts with travel agent system to serve users. It employs the Chinese speech recognition engine supported by Hidden Markov Models technology and IBM ViaVoice Chinese Text-to-Speech engine and Microsoft Agent altogether providing the speaking and hearing abilities and furthermore the visual animations. Result from the limitation of speech recognition engine, sometimes, it may get wrong hearing; besides, the recognition response time is enormously correlated with the amount of words to be recognized. By controlling the words set to be recognized applying dialogue strategies, as wrong speech recognition happens it can dialogue with users continuously and smoothly to percept user’s demanded word. We are in particularly interested in the process of wrong speech recognition of general words. We solve the problem that there is limitation in the amount of words to be recognized. For the condition that word sets have few observation characters and the condition that users mistake speaking and the case to distinguish between homophones, we can still find out the correct users’ demanded words. We can also realize whether a word is in the valid word set. Moreover, by dialogue history records, users can restart their not-yet set-up dialogue items. For those who are not familiar with the domain knowledge of service systems, they could invite the dialogue interface agent to guide them to query related information.
|
|---|
