Sixth International Conference on Spoken Language Processing
October 16-20, 2000
Speaker Information Enhancement
Eloquent Technology Inc.
Ithaca, NY, USA
This study consists of two experiments*. Experiment A
investigates speaker-information distribution in the parametric
domain. Experiment B compares different weighting strategies
for speaker-information enhancement. The results indicate that
weightings based on speaker-information distribution in the
parametric domain yield better speaker recognition performance.
how it statistically affects the separation of this speaker from the
rest speakers in the database. The degree of its effect can be
used as an indicator of the amount of speaker-information coded
in that particular MFCC coefficient.
Chen, Fangxin (2000):
"Speaker information enhancement",
In ICSLP-2000, vol.1, 758-761.