ISCA Archive Eurospeech 1993
ISCA Archive Eurospeech 1993

Generation of speech reply in the speech response system

Keikichi Hirose, Yasuharu Asano

A method was developed for generating speech responses, and was realized on a computer as a subsystem for the speech response system of questions and answers. This subsystem consists of two major parts; those for dialogue processing and dialogue speech synthesis. The former part manages the dialogue flow and generates the content of response in the form of semantic representation. In order to increase the adaptability to dialogue topics, topic-dependent rules of example-based description were introduced. The latter part first generates surface sentences for the response content. In order to increase the processing efficiency and the quality of output speech, the surface sentences are represented by phonological symbols which may have direct correspondence with the pronunciation for the sentences. The response speech is then generated using the speech synthesizer basically developed for the text-to-speech conversion system. Experiments were conducted on the response speech generation for the topic of "ski areas" and the results showed the validity of the developed method.

Keywords: Dialogue Processing, Response Content Generation, Speech Synthesis, Prosodic Features, Focus