The AT&T LVCSR-2000 System (2000)
| Citations: | 5 - 2 self |
BibTeX
@MISC{Ljolje00theat&t,
author = {Andrej Ljolje and Donald M. Hindle and Michael D. Riley and Richard W. Sproat},
title = {The AT&T LVCSR-2000 System},
year = {2000}
}
Bookmark
 |
 |
 |
 |
 |
 |
OpenURL
Abstract
We describe the AT&T recognition system used in the Large Vocabulary Conversational Speech Recognition (LVCSR-2000) evaluation. It is based on multi-pass rescoring of weighted Finite State Machines (FSMs) using progressively more accurate models. The final stages, which provide highest recognition accuracy, use combined outputs of several models that were designed so that they are as different from each other as possible. Acoustic models used in the system varied along three different dimensions aimed at providing maximally different acoustic models: gender dependency, cepstral variance normalization and triphone vs. pentaphone. We used six out of the total of eight possible combinations of model features. The last steps use the outputs from all the models to improve the Maximum Likelihood Linear Regression (MLLR) adaptation step and finally the recognition results with the adapted models are all used to provide the single best combined recognition hypothesis. 1. INTRODUCTION Large Vo...
Citations
No citations identified.
