Skip to main content

Table 2 Evolution of CER and acoustic-model size according to the number of Gaussian components per state (context-free models). The acoustic vectors are composed of 39 coefficients (12 PLP plus energy with and ). 11 136 tests performed on the corpus VODIS (voice command/noisy).

From: Compact Acoustic Models for Embedded Speech Recognition

No. of gauss/state No. of parameters CER
2 17 064 5,48%
4 34 128 3,40%
128 1 092 096 1,80%