How to improve recognition time in CMU Sphinx?

Question

linux

It is necessary to add voice recognition to the asterisk. We decided to start with an offline solution. Based on CMU Sphinx. However, his work made us very sad. The recognition time for simple phrases with a limited vocabulary (3-5 words) takes about 20 seconds, which is not acceptable. A logical question arose: is it that our hands do not grow from there, or is this a limitation of the Sphinx. Recognition was done with the following command:

pocketsphinx_continuous \
    -samprate 8000 \
    -dict my.dic \
    -lm ru.lm \
    -hmm zero_ru.cd_cont_4000 \
    -maxhmmpf 3000\
    -maxwpf 5\
    -topn 2\
    -ds 2\
    -logfn log.log \
    -remove_noise no \
    -infile 1.wav

Which was taken from the Russian language model as a basis. Changing the acoustic model does not affect the picture. Is it worth it and how to dig the Sphinx, or is it useless and you have to go to cloud solutions?