INTERSPEECH 2004 - ICSLP
8th International Conference on Spoken Language Processing

Jeju Island, Korea
October 4-8, 2004

Automatic Network Optimization of Voice Applications

Juan Huerta (1), Chaitanya Ekanadham (2)

(1) IBM T.J. Watson Research Center, USA
(2) Stanford University, USA

We present a technique to automatically transform a voice application's call flow, or interaction graph, into the set of application pages that encapsulate the call flow nodes and resources needed to run such application in a client-server environment. Our technique, called NOVA, performs this automatic partition by making use of a set of cost functions that model the latencies associated with the transmission of data and application resources through a network and their processing by the client components (e.g., voice browsers, grammar compilers and system engines). Our technique facilitates the existence of tools that permit application developers to focus on designing the call-flow of the application while leaving the task of segmenting and packaging the application into pages to our algorithm. The cost functions utilized by our algorithm can be dynamically computed allowing for runtime application optimization. We demonstrate the impact of our algorithm through simulation experiments.

Full Paper

Bibliographic reference.  Huerta, Juan / Ekanadham, Chaitanya (2004): "Automatic network optimization of voice applications", In INTERSPEECH-2004, 2617-2620.