basic-training.sh

#!/bin/bash
# Obtain the latest version of the CMU dict from github
wget https://raw.githubusercontent.com/cmusphinx/cmudict/master/cmudict.dict -O cmudict.dict

# Get rid of the alternative pronunciation markers and normalize formatting
cat cmudict.dict | perl -e 'while (<>) {
                         chomp;
                         @_ = split (/\s+/);
                         $w = shift (@_);
                         $w =~ s/\([0-9]+\)//;
                         print $w."\t".join (" ", @_)."\n";
                       }' \
                             > cmudict.formatted.txt
#Align the dictionary
phonetisaurus-align --input=cmudict.formatted.txt \
    --ofile=cmudict.formatted.corpus \
    --seq1_del=false

#Train an n-gram model with mitlm (or whatever LM training toolkit you like
estimate-ngram -o 8 -t cmudict.formatted.corpus -wl cmudict.formatted.o8.arpa

#Transform into an FST
phonetisaurus-arpa2wfst-omega --lm=cmudict.formatted.o8.arpa \
    --ofile=cmudict.formatted.o8.fst

#Test with some word 'TESTING'
phonetisaurus-rex.py --model cmudict.formatted.o8.fst \
    --word testing