Hybrid in-car speech recognition for mobile multimedia applications

This paper presents a speech recognition system that is particularly suited for drivers and passengers in a vehicle environment. The requirements of such a system are quite unique. Some applications require small vocabulary with a quick response time while others demand a large vocabulary, however, they can sustain a larger latency in response. The idea of our approach is to complement the scope of an in-car speech recognizer with a large vocabulary recognizer running on a server using a client/server architecture without increasing the overall costs. Apart from the recognizing concept, the paper describes a hybrid speech interface (HSI), in which the back-end recognizer is embedded. The HSI architecture is an open architecture for in-car clients and their underlying infrastructure. It allows the integration of different applications (NewsReader, Route Guidance, etc.) into a homogeneous user interface