Skip to main content

(b) ULT+WRE approach

From: Compact Acoustic Models for Embedded Speech Recognition

  Without adaptation With adaptation
  MLE FD MLE FD
Very compact model (6 k.) 5,25% 5,11% 4,76% 4,48%
Compact model (11 k.) 4,01% 4,27% 3,64% 3,80%