Digital Signal Processing Reference
In-Depth Information
4.
CONCLUSIONS AND FUTURE WORK
This work represents a preliminary step in the development of a dialogue
system for in-car voice interaction with advanced services of navigation
assistance and tourist information access. As the system complexity in the
given framework is a crucial aspect, our research focuses on the development
of multiple fast recognition units and a suitable combination strategy that may
lead to better performance than the adoption of a single full-coverage
recognizer.
Even if the application of a maximum likelihood criterion to select the
recognition output represents the simplest choice, it offers some advantages in
terms of performance and also complexity, assuming the ability of the
dialogue manager to predict at each interaction step the most likely domain in
terms of geographic area as well as of dialogue context. The geographic
clustering seems to be effective in presence of large list of names (cities,
streets and hotels names) that give rise to a considerable acoustic
confusability. Work is under way for what regards the selection of more
reliable outputs, on the basis of confidence measures and word hypotheses
graphs.
REFERENCES
[1]
Proceedings of the Hands-Free Speech Communication Workshop (HSC), Kyoto (Japan),
2001.
M. Omologo, P. Svaizer, M. Matassoni, “Environmental conditions and acoustic
transduction in hands-free speech recognition”, Speech Communication, vol.25, pp. 75-
95, 1998.
[2]
