Toward On-Line Learning of Chinese Continuous Speech Recognition System.

Rong Zheng,Zuoying Wang
DOI: https://doi.org/10.21437/icslp.1998-748
1998-01-01
Abstract:In this paper, we presented an integrated on-line learning scheme, which combined the state-of-art speaker normalization and adaptation techniques to improve the performance of our large vocabulary Chinese continuous speech recognition (CSR)system. We used VTLN to remove inter-speaker variation in both training and testing stage. To facilitate dynamic transformation scale determination, we devised a tree-based transformation method as the key component of our incrementaladaptation. Experiments shows that the combined scheme of on-line learning (incremental & unsupervised) system, which gives approximately 22~26% error reduction rate, was proved to be better than either method when used separately at and 2.7 . .
What problem does this paper attempt to address?