Skip to main content

Table 1 Word error rate (%)and the number of parameters of the proposed DTAM systems and conventional GMM systems on the 1992 WSJ non-verbalized 5K closed-test set

From: Decision tree-based acoustic models for speech recognition

System

% WER

Number of parameters

DTAM monophone

22.9

451k

DTAM triphone

12.9

766k

DTAM triphone with gender information

11.9

770k

GMM triphone 4 components/mixture

10.1

870k

GMM triphone 8 components/mixture

7.5

1740k