The application of DLB algorithm in distributed speech dialogue system

This paper presents our recent work towards development of a distributed speech interactive system, which was a human computer interactive dialogue system for stock and calling for people by phone. This system integrated speech recognition (ASR), text to speech (TTS) and natural language perception (NLP) technologies with a distributed architecture to allow multi-users to access and converse with it concurrently. During a typical telephone based interaction, users could retrieve up-to-date stock information and calling for somebody by directly speaking their name. Our paper proposed an efficient algorithm with sender active and non-preemptive concepts in order to solve the loads allocating when multi-users accessed the system and gave rise to much burden to the recognition servers and TTS servers. This paper was composed of five parts, 1) an introduction about why such architecture was built; 2) an overview of the system; 3) proposal and implementation of the algorithm; 4) evaluation and measurement of the system performance based on the architecture; and 5) gives a brief summary.