Constructing a Prototype Spoken Dialogue System for World Wide Named Places

Abstract

Named places, such as named restaurants and gas stations, are usually not supported in spoken dialogue systems used in vehicles. They usually support navigation in an invehicle infotainment system in the vehicle, but the user cannot mention named places. This thesis attempts to construct a prototype dialogue system using an issue-based dialogue manager. The idea is to support named places across the world, in the context of driving a vehicle. The dialogue system is integrated with an in-vehicle infotainment system to enable spoken dialogue control of navigation and weather use cases. The spoken dialogue system is grammar based and all the places used are loaded on startup. To enable world wide named places, new places are loaded dynamically at run-time, fetched from the cloud. This solution however introduces shortcomings, for example a long delay when waiting for the speech recognition grammar to compile. This problem also showed during user tests, being pointed out as one of the biggest problems with the prototype. Users also compared it to world-leading competition and pointed out the di erences. The most important di erences, on a conceptual level, can be solved by switching the grammar based speech recognizer for a freeform one. This solution would be interesting future work.