Institute of Information Technology
- 1:
About the Institute. - 2: Staff.
- 2.1:
Information Transmission. - 2.2: Dialogue Systems.
- 2.2.1:
Wolfgang Minker . - 2.2.2: Gregor Bertrand.
- 2.2.3: Tatiana Gasanova .
- 2.2.4: Tobias Heinroth .
- 2.2.5: Helmut Lang.
- 2.2.6: Florian Nothdurft.
- 2.2.7: Alexander Schmitt.
- 2.2.8: Stefan Ultes.
- 2.2.9: Kseniya Zablotskaya.
- 2.2.10: Sergey Zablotskiy.
- 2.2.11: Cristina Rico Garcia.
- 2.2.12: Hansjörg Hofmann.
- 2.2.13: Dmitry Zaykovskiy.
- 2.2.1:
- 2.3: Alumni.
- 2.1:
- 3: Teaching.
- 4: Information Transmission.
- 5: Dialogue Systems.
- 6: Intranet.
- 7: Links.
Optimising Client-Server Architectures for Spoken Language Dialogue Systems
Author: Dmitry Zaykovskiy
Status: in progress
Description:
The subject of this thesis is centred around optimising spoken language dialogue systems (SLDSs) architectures. The study builds upon previous work in the areas of speech signal coding, information transmission, automatic speech recognition and spoken natural language understanding. It aims at improving both the performance of the speech processing components and the upstream data transmission channel. The topic is motivated by an increasing use of the SLDSs technology in all (especially mobile) situations of life and for increasingly resource-consuming speaker-independent and large-vocabulary applications. This implies that the main processing steps of SLDSs are shifted from the client-side (e.g. Personal Digital Assistants - PDAs and mobile phones) to a powerful server computer. However, such a paradigm shift requires investigating data pre-processing, transmission and speech processing techniques that are optimally and mutually tuned.
In order to achieve the goal of an optimum overall SLDSs architecture, an experimental set-up containing a simulated speech data transmission channel front-end and a speech processing back-end including components for stochastically-based speech recognition and spoken natural language understanding are established. The major task of the thesis work consists of adapting and optimising the individual component approaches of the set-up so as to obtain an optimum overall system performance in terms of high accuracy and low transmission costs.
The utility of the developed approaches is illustrated and evaluated with real user data on applications for information retrieval and transaction services. The thesis is carried out in cooperation with the Information Transmission Group of the ITUU Department.
PhD_Description.pdf